Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The International Association for the Study of Lung Cancer Early Lung Imaging Confederation

The International Association for the Study of Lung Cancer Early Lung Imaging Confederation review articles abstract The International Association for the Study of Lung Cancer Early Lung Imaging Confederation 1 2 3 4 5 James L. Mulshine, MD ; Ricardo S. Avila, MS ; Ed Conley, PhD ; Anand Devaraj, MD ; Laurie Fenton Ambrose, BS ; 6 7 8 9 10 Tanya Flanagan, MA ; Claudia I. Henschke, MD, PhD ; Fred R. Hirsch, MD, PhD ; Robert Janz, MSci ; Ryutaro Kakinuma, MD, PhD ; 11 12 13 13 14 Stephen Lam, MD ; Annette McWilliams, MBBS ; Peter M.A. Van Ooijen, PhD ; Matthijs Oudkerk, MD, PhD ; Ugo Pastorino, MD ; 15 16 16 17 13 18 Anthony Reeves, PhD ; Patrick Rogalla, MD ; Heidi Schmidt, MD ; Daniel C. Sullivan, MD ; Haije H.J. Wind, MSc ; Ning Wu, MD ; 19 20 7 3 Murry Wynes, PhD ; Xie Xueqian, MD, PhD ; David F. Yankelevitz, MD ; and John K. Field, PhD PURPOSE To improve outcomes for lung cancer through low-dose computed tomography (LDCT) early lung cancer detection. The International Association for the Study of Lung Cancer is developing the Early Lung Imaging Confederation (ELIC) to serve as an open-source, international, universally accessible environment to analyze large collections of quality-controlled LDCT images and associated biomedical data for research and routine screening care. METHODS ELIC is an international confederation that allows access to efficiently analyze large numbers of high- quality computed tomography (CT) images with associated de-identified clinical information without moving primary imaging/clinical or imaging data from its local or regional site of origin. Rather, ELIC uses a cloud-based infrastructure to distribute analysis tools to the local site of the stored imaging and clinical data, thereby allowing for research and quality studies to proceed in a vendor-neutral, collaborative environment. ELIC’s hub-and- spoke architecture will be deployed to permit analysis of CT images and associated data in a secure envi- ronment, without any requirement to reveal the data itself (ie, privacy protecting). Identifiable data remain under local control, so the resulting environment complies with national regulations and mitigates against privacy or data disclosure risk. RESULTS The goal of pilot experiments is to connect image collections of LDCT scans that can be accurately analyzed in a fashion to support a global network using methodologies that can be readily scaled to accrued databases of sufficient size to develop and validate robust quantitative imaging tools. CONCLUSION This initiative can rapidly accelerate improvements to the multidisciplinary management of early, curable lung cancer and other major thoracic diseases (eg, coronary artery disease and chronic obstructive pulmonary disease) visualized on a screening LDCT scan. The addition of a facile, quantitative CT scanner image quality conformance process is a unique step toward improving the reliability of clinical decision support with CT screening worldwide. JCO Clin Cancer Inform 4:89-99. © 2020 by American Society of Clinical Oncology Licensed under the Creative Commons Attribution 4.0 License INTRODUCTION tomography (CT) screening. The concept of de- veloping an early lung cancer image registry emerged Lung cancer is the most lethal cancer throughout the ASSOCIATED through a series of workshops sponsored by the In- CONTENT world, and it typically presents at a late stage when ternational Association for the Study of Lung Cancer 1,2 Data Supplement cure is unlikely. Recent reports have demonstrated (IASLC) as a critical opportunity to accelerate the pace Author affiliations the usefulness of low-dose computed tomography (CT) of innovation in improving the curative management and support screening in reducing lung cancer mortality in heavily information (if 3-7 for detection and intervention with early lung cancer. tobacco-exposed individuals. However, to achieve applicable) appear at the most efficient screening management, groups The IASLC mission is to improve lung cancer outcomes the end of this have incorporated a quantitative assessment of pul- through international and multidisciplinary collabora- article. monary nodule volume to guide the diagnostic tive efforts. This large collaborative image archival Accepted on 8-12 December 5, 2019 case–finding efforts within the screening process. and analysis effort builds on the established IASLC and published at In this fashion, the false-positive detection rate can be successes with the national and international lung ascopubs.org/journal/ 10-12 reduced from 28% to approximately 3%. This cancer IASLC TNM Staging Committee, which has cci on February 6, improves the screening cost and reduces the potential been undertaken in collaboration with the Union for 2020: DOI https://doi. for iatrogenic harm, which would be a critical aspect to International Cancer Control and American Joint org/10.1200/CCI.19. 00099 include in the global implementation of computed Committee on Cancer, as well as with the recently 89 Mulshine et al CONTEXT Key Objective Can an open-source cloud-based environment become a repository of screening computed tomography images and as- sociated data to enable quantitative and related imaging tool development for use in guiding the management of early, presymptomatic lung cancer and related thoracic diseases? Knowledge Generated A prototype international, open-source imaging resource is proposed that can allow federated image/data interrogation. This construct is designed to comply with existing international standards for data security while enabling the development and validation of new imaging biomarkers to facilitate early lung cancer management. Relevance Lung cancer screening is emerging as an important approach for early lung cancer management. However, robust and economic image analysis tools are needed to ensure facile clinical workflows for pulmonary nodule detection and quantitative nodule assessment so that this new service can be provided to the target high-risk population at high quality throughout the world. updated lung cancer pathology collaboration with the all of the images and metadata within their defined spoke 14,15 WHO. Success in developing screening tools, as with environment consistent with local governing data-sharing the staging and pathology efforts, involves issues of scale provisions. With local site permission, their stored clinical and cost that leverage the IASLC’s broad international and imaging data can be made accessible, to allow soft- scope and expertise in aligning global participation to ware tools distributed to the spoke from the hub to the improve early lung cancer management. relevant stored digital data. In this confederated architec- ture, the hub acts as a conduit with the spokes to dis- To date, no consensus image analysis tool has emerged to tribute software analysis tools to the relevant spokes and allow routine and reliable volumetric characterization of then to aggregate the results of the analysis of the images pulmonary nodules in routine clinical imaging settings. stored locally in the participating spokes. Therefore, only Developing a tool to allow easy and robust nodule mea- the resulting analysis data will leave an individual spoke. surement requires access to large numbers of high-quality The resulting analysis data aggregated in the hub from thoracic CT images that were acquired with the intent of multiple sites with diverse populations allow for research precisely measuring volumes of pulmonary nodules 5 to and for quality questions to be addressed with a poten- 10 mm in diameter. Because this is a newly appreciated tially vast number of test screening cases from multiple opportunity as lung imaging technology rapidly improves, countries. lung images from earlier CT screening trials were generally not acquired with sufficient resolution to enable this pre- To accommodate this architecture, the most cost-feasible cise volumetric tool development. Therefore, prospective approach for a flexible, scalable, and sustainable envi- collections of CT images from current-generation, high- ronment capable of enabling the goals of ELIC is to leverage resolution CT scanners are urgently required to address the global accessibility of a cloud environment. Currently, thoracic CT screening images and associated clinical this gap. This lack of large quantities of such high-quality outcomes and relevant metadata are stored in a vast array image data imposes a profound barrier to progress with of architectures across IASLC member sites. We envision early lung cancer management. developing a vendor-neutral, secure, scalable, cloud-based How to Bring Value to Early Lung Cancer Detection environment to bridge to existing sites’ data storage re- In response to this situation, the IASLC hosted a planning sources. Table 1 summarizes the design considerations workshop held in Dallas, Texas, in February 2018. Al- guiding the development of this informatics resource. Given though aware of the heterogeneous nature of existing image the dynamic and complex nature of the privacy challenges registries at leading centers from around the world, the inherent in collecting and sharing large amounts of imaging group proposed the creation of a cloud-based informatics and clinical data, the proposed IASLC imaging/data re- infrastructure to interact with existing international regis- source may be preferable for many national sponsors to tries and centers collecting thoracic CT images together have a rigorously designed, precompetitive environment with associated core clinical outcomes data to optimize cost hosted by an international, nonprofit professional society and data security. such as IASLC. IASLC has a proven legacy of patient This IASLC Early Lung Imaging Confederation (ELIC) was benefit and as a reliable host to ensure appropriate proposed as a hub-and-spoke architecture with the in- stewardship as an “honest broker” for such a critical in- 14,15 tention of enabling the imaging-donating local site to retain ternational resource. 90 © 2020 by American Society of Clinical Oncology IASLC-ELIC Initiative TABLE 1. Guiding Principles of ELIC acquisition protocols for CT lung nodule measurement. The data are also revealing new insights into low-dose CT lung In alignment with its mission and charge, ELIC seeks to share its resource with both cancer screening image acquisition performance, in- national and international scientific lung cancer research communities by: cluding the reproducibility of image acquisition properties, Promoting the common good by maximizing collaborative research for the differences between requested and obtained slice thick- benefit of all ness, and losses of resolution associated with lowering the Making ELIC research data available to the research community to advance radiation dose. scientific knowledge Providing open-source resources including an environment and tools to METHODS ensure broadest dissemination of resources to enhance outcomes with A proof-of-concept demonstration was developed in 2018 tobacco-related diseases to evaluate the potential of a hub-and-spoke–distributed Ensuring the generation of high-quality research lung cancer screening image archive and computing ar- Respecting the legal rights and legitimate interests of all stakeholders involved chitecture to achieve the goals of ELIC . Ten international Protecting the privacy of its research participants and the confidentiality of cloud computing sites were chosen from the Amazon Web their data Services global cloud network map to launch and set-up Promoting transparency, responsibility, interoperability, and fairness spoke EC2 cloud computing instances, as shown in Figure Ensuring accountability and oversight 1. This figure overlays the 10 ELIC spokes feeding a central hub (at the point of the arrow) on the Amazon Web Services Enriching the content of the ELIC database, including through the return of high quality–derived data by researchers (AWS) global cloud network map. The hub is shown (with the blue circle at the tip of the arrow), with each of the 10 spokes Managing access to ELIC data to balance current and future needs (indicated by green circles) populated with an identical set of Abbreviation: ELIC, Early Lung Imaging Confederation. 100 publicly available de-identified CT lung scans. As shown on the map, existing AWS cloud services sites are indicated by smaller (lighter) blue and purple circles. This distribution High-Quality Image Acquisition of existing cloud resources demonstrate the global reach of For optimal efficiency, there is also the need to pro- AWS to support local/regional hosting of available lung spectively ensure that newly accrued thoracic CT images cancer screening images and corresponding clinical data as are objectively of sufficiently high quality to support robust required by General Data Protection Regulations. Each of quantitative analysis. Because ELIC is envisioned to be these spokes was populated with an identical set of 100 a resource to develop and then validate image processing publicly available de-identified CT lung scans. However, algorithms as required to support clinical decision making each scan was given a unique patient ID and fictitious age, for early lung cancer management, such efforts will move sex, and pack-year demographics. This was done so that into the rapidly evolving realm of deep learning methods or proof-of-concept performance testing could be obtained with artificial intelligence approaches; therefore, the curation of the equivalent of 1,000 patient cases; the analyses de- an image collection that has been optimized for quantitative scribed in this report were repeated on at least 5 separate 17,18 measurement is critical. occasions. In addition, a Hub EC2 cloud computing instance Reliable guidance and clinical management in the lung was set up at the Northern Virginia location of the Amazon cancer screening setting requires accurate and robust Web Services cloud. analysis of pulmonary nodule volume. Therefore, ELIC will Figure 2 is a schematic representation of the computing require image-sharing sites to prospectively adhere to the resources and main data flows that were used during the Quantitative Imaging Biomarkers Alliance (QIBA) of the design of the ELIC Hub-and-Spoke Environment (H&SE) Radiologic Society of North America (RSNA) CT Small Lung software. A lung cancer screening principal investigator Nodule Profile quality provisions to ensure that aggregated (PI; A) typically coordinates CT lung cancer screening imaging data obtained for quantitative analysis are of a high for several screening sites shown as sources (Ai). The PI 19,20 quality; this is a unique strength of the ELIC effort. is responsible for submitting de-identified lung cancer From extensive work in optimizing image measurement screening image DICOM (Digital Imaging and Communi- quality in this setting, the QIBA has developed a process to cations in Medicine) data and metadata, including any 17,18 address these factors. This entails analyzing pulmonary requested image annotations, before data upload to a nodules in the volume range relevant to lung cancer spoke (B). Each time data on the spoke (B) are added or screening (6-10 mm in diameter). Crowd-sourced data corrected, a set of basic demographics and metadata in- collected from international QIBA CT Small Lung Nodule formation about the new data is communicated to the hub testing sites have been used over the past 2 years to help (C). In this way, the hub and spoke will remain in agreement identify and globally disseminate the top-performing CT regarding the data that a PI has made available for analysis image acquisition protocols for each CT scanner model. on the spoke (B). Because the spoke (B) can be set up on This innovative approach has enabled what we believe to a local cloud computing instance or on local computing be the first rapid global optimization of scanner image hardware, the spoke (B) data will remain within the source JCO Clinical Cancer Informatics 91 Mulshine et al Phantom image quality testing Edge locations Multiple edge locations Regional edge caches FIG 1. The global distribution of the hub and spokes for the 2018 World Conference on Lung Cancer ELIC proof-of-concept demonstrations using the Amazon Web Services (AWS) cloud. geographic region and will only be analyzed according to contain fully automated scripts for thoroughly performing strictly administered analysis and reporting rules (D) de- required data protection operations, such as the General cided by the PI (A). The data residing on all spokes are de- Data Protection Regulation “right to be forgotten.” Each identified by the PI (A) before upload. ELIC spoke will have automated scripts that achieve the data governance protections and procedures required by Thus, at all times, only de-identified data are stored on the local regulations, some of which will be standard and can ELIC H&SE and, at rest, the data are also encrypted. Each be provided by the same ELIC H&SE script for all regions. ELIC spoke contains a de-identified set of data stored on a spoke using identical ELIC H&SE data storage organi- A client (E) can view a listing or catalog of all the data sets zation and methods. This allows the ELIC H&SE software to available within the ELIC environment and take the Source ELIC Environment Screening Source A PI Source Client Cloud Basic Meta Instance Hub Spoke Algor gorithm Cloud Instance Res esul u ts FIG 2. High-level illustration of the ELIC Hub & Spoke Environment and how lung cancer screening principal in- vestigators (PI) and clients will interact with the system. DICOM, Digital Imaging and Communications in Medicine; ELIC, Early Lung Imaging Confederation. 92 © 2020 by American Society of Clinical Oncology Results Algorithm Basic Meta DICOM Meta Data IASLC-ELIC Initiative necessary steps to run a quantitative algorithm on all the the quality of a CT scanner and acquisition process. The data or on a subset of the data available within the ELIC results of the comprehensive QIBA CT image quality environment. This is referred to as running an “experiment” analysis are sent back to the site within a few minutes using within the ELIC environment. To run an experiment, the an easy-to-interpret structured report indicating whether client (E) must provide a quantitative algorithm according to the CT scanner and acquisition parameters were of suffi- ELIC H&SE specifications to the hub (C) and request that cient quality for quantitative assessment. If the image the hub execute the experiment on all the relevant spokes quality was insufficient, then remediation measures are (B). The hub (C) then coordinates the running of the suggested to the site to improve image quality. This analysis quantitative algorithm on the spokes and collects the re- has been performed at. 60 CT lung cancer screening sites sults according to the analysis rules (D) agreed to by the PI. evaluating the use of the CTLX1 phantom, including When the experiment is finished, the client (E) is provided screening sites in Australia, England, Canada, China, Israel, an aggregate summary of all the data collected from the Italy, the Netherlands, Poland, Spain, Japan, and the experiment, including information on any errors or issues United States. The image quality data collected from the encountered while running the experiment. A prototype analysis of CTLX1 phantom scans include CT scanner hub password-protected Website was created that dis- image acquisition parameter settings such as milliampere, plays the status of spokes, allows the specification and peak kilovoltage, slice thickness, slice spacing, and re- construction kernel, as well as fundamental image quality launching of quantitative lung imaging experiments on characteristics achieved, including levels of edge en- global collections of data, and provides a real-time quan- hancement, 3-dimensional (3D) resolution, 3D resolution titative and graphic display of the results obtained from the aspect ratio, CT linearity, noise, and 3D spatial warping. international spoke calculations. Each of these fundamental image quality properties is Two cloud-based experiments using software algorithms measured throughout the CT scanner field of view at 3 were created to accept DICOM data and produce quanti- distances from isocenter (0, 100, and 200 mm) to ensure tative results and images in a standard format that the ELIC that lung nodules present in the lung periphery, which is H&SE supports. One algorithm quantitatively measured common, can be accurately measured. total lung volume, and the second algorithm quantitatively For proof-of-concept testing, the QIBA CT Small Lung measured the volume of solid pulmonary nodules. This Nodule Profile Conformance automated phantom analysis permitted the ELIC H&SE to demonstrate running the software was placed on SPOKE 1 running in northern quantitative imaging experiments on image collections Virginia and used to run automated image quality analyses at globally distributed spokes and then aggregating the on scans of the QIBA CT Small Lung Nodule Profile CTLX1 quantitative CT image measurements and output image phantom. These tests confirmed that the fully automated results on the hub. Each of the 10 spokes was set up to run QIBA CT Small Lung Nodule Profile conformance certifi- one of the 2 quantitative CT lung image measurement cation methods for CT image quality assessment will be algorithms when requested by the hub. able to successfully run on future ELIC spokes (or hub) The project also deployed the QIBA CT Small Lung Nodule running on the Amazon cloud. Profile Conformance Certification service on the hub and A total of 5 live demonstrations were run using the ELIC performed CT image quality conformance assessment environment at 5 distinct time windows to evaluate the calculations on CT phantom scans. Cloud-based phantom ability of ELIC H&SE to perform useful quantitative imaging analysis software was developed to perform a low-cost CT computational experiments on large collections of globally image quality assessment using a specifically designed distributed CT lung cancer imaging cases without moving phantom (ie, test object). This approach makes achieving data out of geographic regions. Each demonstration was CT image quality conformance with the QIBA Small Lung run for ELIC H&SE analyses performed with a central hub Nodule Profile possible from virtually any clinical imaging 17,18 server and 10 globally distributed spoke servers all running site in the world. The purpose of this quality control step on the Amazon Web Services cloud. The prototype ELIC H& is to ensure consistent image quality appropriate for the SE Website allowed a user to launch a computational defined context of use with lung cancer screening–related experiment request on a specific set of data distributed quantitative assessment for imaging sites around the world. across any number of spokes, store the results, and display The standardized assessment of CT lung imaging quality is an aggregated summary report of the results when the enabled using a dedicated, low-cost phantom (CTLX1) that analysis was complete. was developed for this purpose. The CTLX1 phantom contains small, precision-made, geometric components to In all, 3 pilot testing runs were performed. For these pilot assess thoracic CT imaging performance. A CT scan of the testing runs, quantitative lung volume analyses were CTLX1 phantom is typically acquired in approximately completed on 1,000 globally distributed CT lung cancer 5 minutes and then uploaded to the QIBA Phantom screening data sets in , 25 minutes at 10 different ELIC Analysis Service, which rapidly analyzes the uploaded hosting sites from 4 continents. Figure 3 shows the main phantom image using automated software to characterize user page available within the Website, which includes JCO Clinical Cancer Informatics 93 Mulshine et al FIG 3. The main user page available at www.iaslc-elic.org for the launch and review of quantitative lung imaging experiments on globally distributed spokes. a listing of the available spokes, a place to specify the software is set up so that each spoke performs a compu- launch of a quantitative experiment on the data located on tational request and returns quantitative results back to the the spokes, and a list of pages showing the results obtained hub. Figure 4 shows the results page from a representative from each experiment. lung nodule measurement experiment (experiment 3) performed on all the images at the 10 sites in the archive The developed H&SE software allows a hub server to make with lung nodules (N = 62 × 10) in the ELIC H&SE. The quantitative CT imaging computational requests to a col- results reported were mean volume and standard deviation lection of globally distributed spokes, each of which is populated with de-identified CT lung images. The H&SE for all the data sets analyzed. Figure 5 displays the detailed 94 © 2020 by American Society of Clinical Oncology IASLC-ELIC Initiative FIG 4. Results page for a lung nodule measurement experiment performed in the ELIC Hub and Spoke Experiment. quantitative imaging results that were computed on CT lung Toronto, Canada. In total, . 5 live demonstrations suc- imaging cases on SPOKE 10 (Sao Paulo, Brazil) for the cessfully showed the simultaneous running of the CT lung nodule algorithm on CT lung imaging cases distributed at same experiment as that shown in Figure 4 (experiment 3), 10 internationally distributed spoke sites. including the quantitative lung nodule volume values and 3 orthogonal reformat images with contours of the segmen- DISCUSSION tation results overlaid for each case. This initial proof-of-concept project demonstrated the po- Three presentations, including a live demonstration, were tential of the ELIC H&SE to provide a useful resource for given at IASLC’s 2018 World Conference on Lung Cancer in global quantitative lung imaging studies. This project used JCO Clinical Cancer Informatics 95 Mulshine et al FIG 5. Detailed results for cases run on spoke 10 (Sao Paulo) for experiment 13. global cloud-computing resources, each populated with an QIBA CT Small Lung Nodule Profile CT image quality identical data set of 100 publicly available lung cancer conformance testing service was successfully run on a spoke to demonstrate that CT image quality monitoring images. A central hub Website simultaneously distributed and optimization can be supported with the ELIC H&SE. 2 open-source quantitative lung measurement algorithm requests to the 10 globally distributed spoke sites. Using Although the developed proof-of-concept software imple- publicly available CT lung images allowed the project to mented a basic H&SE, the functionality represents a subset achieve results quickly. The hub received and aggregated of that which will be needed to fully realize the IASLC ELIC all quantitative algorithm results from the spoke sites and H&SE vision. For example, the demonstration Website displayed the results on the hub Website. In addition, the supports only 1 user type; when mature, the ELIC H&SE 96 © 2020 by American Society of Clinical Oncology IASLC-ELIC Initiative will support a variety of different roles and types of access of the processes for the users of the spokes and hub re- for clients, hub personnel, and spoke personnel. Future quires a well-functioning informatics environment with versions of the ELIC H&SE will also provide higher levels of easy-to-setup and deployment tools that will enable rapid security to prevent access or extraction of PI CT image data screening care implementation as well as research by new as well as to prevent unauthorized use of client algorithms. global lung cancer screening groups. For these initial demonstrations, the client algorithms were The power of ELIC relates to the accessibility of the cloud created as open-source projects and built on each spoke and its remarkable cost efficiency. For image quality into an executable application. However, a future ELIC H& processes, ELIC already uses machine vision and will soon SE will likely also support the distribution of algorithms from include artificial intelligence to ensure optimal and eco- the hub. For example, the distribution of virtual containers nomic image quality. This resource can greatly accelerate with executable applications embedded is being explored. radiomics and deep learning processes for medical images Future H&SE environmentswillalsoneedtoprovideadditional and can be integrated with digital pathology and genomic functionality andsupportforartificialintelligencedevelopment data. As the ELIC project continues to mature as a resource and specifically, deep learning algorithm development and to conduct analyses and study international collections of testing. A fuller description of the methodology, governance, high-quality thoracic CT images together with associated data use provisions, scope, and other proposed policy ele- biomedical data, there are a number of models through ments are provided in the Data Supplement. which the pharmaceutical industry can be involved. For The ELIC H&SE prototype vehicle was developed with example, drug company sponsors may host their own ELIC IASLC seed funding to support feasibility testing and was spoke, on which they can collect quality-controlled CT developed as an open-science and open-source research/ images associated with new innovative neoadjuvant stud- care environment to allow a broad range of collaborative ies, adjuvant or even chemoprevention clinical trials, and participation in achieving its CT imaging-related objectives. associated clinical outcomes data. With each trial, the Funding to sustain this large-scale international collabo- sponsor can decide which image collections to acquire and rative effort will be dependent on the early lung cancer maintain as private versus aggregated in large public image research and innovation communities using ELIC in their collections. In this fashion, aggregated image and data research to address important research and lung cancer collections could emerge as critical postmarketing research care issues. We expect many stakeholders, including and quality resources. Furthermore, access to large quality- software developers, artificial intelligence companies, im- controlled CT screening images with clinical outcomes data aging researchers, diagnostic device developers, medical will provide the basis on which to construct the next imaging companies, government agencies, pharmaceuti- generation of algorithms to build models and tools, which cal developers, academic societies, and many other in- can function as clinical decision support. This is a rapidly novators, to be early users of the flexible capabilities of evolving area, and we are working to evolve ELIC in ELIC. This service model is designed to allow a sustainable alignment with new regulatory guidance relative to re- path forward for this internationally accessible quantitative sponsibly building international clinical decision support 23-25 imaging environment to emerge as a core resource for tools for clinical care and research. Because thoracic improving the detection and management of early thoracic CT images from screening also contain information about disease. the presence of early coronary artery disease and chronic The internationally federated ELIC hub-and-spoke system obstructive disease, a high-quality international collection of these images will be of growing public health value. has features such as robust data privacy provisions to permit large-scale analyses of clinical CT images with In summary, the development and deployment of the ELIC relevant associated data in a secure environment (ie, pri- hub-and-spoke environment, together with fair and inter- vacy protection that is under the control of the Trial/Study nationally developed governance policies, will establish, to PI), so that this resource could support the conduct of our knowledge for the first time, a large and efficient global clinical trials. This approach is intended to ensure local computing environment for the study of thoracic CT scans governance control of the site PI, who can address the obtained in the context of lung cancer screening. The use of specific data protection conditions at diverse international QIBA CT Small Lung Nodule Profile and the associated locations. phantoms and software tools will improve the quality of In addition to prospective accumulation of individual new global thoracic CT images aggregated for ELIC, ensuring screening cases, existing imaging collections will remain at much more efficient image tool development. Not only will the local site where they were collected, so, again, the these resources help accelerate lung imaging research and resulting environment remains consistent with local regu- the availability of thoroughly tested imaging tools, but the lations without a privacy or data disclosure risk. An inherent data generated will provide insights to guide future rec- design feature of ELIC is to manage large numbers of ommendations for lung screening and for managing early thoracic CT image screening registries. Therefore, control thoracic diseases. JCO Clinical Cancer Informatics 97 Mulshine et al Fred R. Hirsch AFFILIATIONS Consulting or Advisory Role: AstraZeneca, Genentech, Lilly, Merck, Rush University, Chicago, IL Bristol-Myers Squibb, Pfizer, Roche, Loxo, Novartis, Amgen, Oncocyte Accumetra, LLC, Clifton Park, NY Research Funding: Amgen (Inst), Merck (Inst), Abbvie (Inst), Biodesix University of Liverpool, Liverpool, United Kingdom (Inst), Mersana (Inst), Rain Therapeutics (Inst) University College, London, United Kingdom Patents, Royalties, Other Intellectual Property: EGFR FISH and IHC for GO2 Foundation, Washington, DC prediction of outcome in patients treated with EGFR inhibitors (Inst) Canadian Partnership Against Cancer, Toronto, Ontario, Canada Icahn School of Medicine at Mount Sinai, New York, NY Stephen Lam Mount Sinai Health System, New York, NY Patents, Royalties, Other Intellectual Property: Deep learning prediction University of Groningen, Groningen, Netherlands algorithm to estimate the 3-year lung cancer risk and cancer related National Cancer Center Hospital, Tokyo, Japan mortality for individuals who have . 2 screening chest CT scans. Joint University of British Columbia, Vancouver, British Columbia, Canada application by Johns Hopkins University and the BC Cancer Agency Fiona Stanley Hospital, Western Australia Patent pending (Inst). University Medical College, Groningen, Netherlands Instituti Tumori, Milan, Italy Annette McWilliams Cornell University, Ithaca, New York, NY Travel, Accommodations, Expenses: Olympus Medical Systems, Roche Toronto Joint Department of Medical Imaging, University of Toronto, Matthijs Oudkerk Ontario, Canada Employment: Institute for Diagnostic Accuracy Duke University Medical Center, Durham, NC Speakers’ Bureau: Siemens Healthineers National Cancer Center, Peking Union Medical College, Beijing, China Research Funding: EU Horizon 2020 (Inst) International Association for the Study of Lung Cancer, Denver, CO 20 Travel, Accommodations, Expenses: AstraZeneca Shanghai General Hospital, Shanghai, China Anthony Reeves CORRESPONDING AUTHOR Leadership: D4Vision James L. Mulshine, MD, Rush University Medical Center, 1700 W Van Stock and Other Ownership Interests: VisionGate Buren St, Suite 245, Triangle Office Building, Chicago, IL 60612; Patents, Royalties, Other Intellectual Property: Co-inventor on patents e-mail: jmulshin@rush.edu. owned by Cornell Research Foundation, which are nonexclusively licensed and are related to technology involving computer-aided diagnostic methods. AUTHOR CONTRIBUTIONS Conception and design: All authors Patrick Rogalla Collection and assembly of data: Ricardo S. Avila Research Funding: Canon Medical (Inst) Data analysis and interpretation: James Mulshine, Ricardo S. Avila, John Travel, Accommodations, Expenses: Canon Medical (Inst) Field David F. Yankelevitz Manuscript writing: All authors Stock and Other Ownership Interests: Accumetra Final approval of manuscript: All authors Consulting or Advisory Role: Grail Accountable for all aspects of the work: All authors Patents, Royalties, Other Intellectual Property: Licensing agreement between Cornell University and General Electric for management of AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF various lung abnormalities including lung nodules. INTEREST John K. Field The following represents disclosure information provided by authors of Consulting or Advisory Role: Epigenomics (Inst), NUCLEIX (Inst), this manuscript. All relationships are considered compensated unless AstraZenca (Inst), AstraZenca (Inst), Institute Diagnostic Accuracy (Inst) otherwise noted. Relationships are self-held unless noted. I = Immediate Speakers’ Bureau: AstraZenca (Inst) Family Member, Inst = My Institution. Relationships may not relate to the Research Funding: Janssen Research & Development (Inst) subject matter of this manuscript. For more information about ASCO’s conflict of interest policy, please refer to www.asco.org/rwc or ascopubs. No other potential conflicts of interest were reported. org/cci/author-center. Open Payments is a public database containing information reported by ACKNOWLEDGMENT companies about payments made to US-licensed physicians (Open The International Association for the Study of Lung Cancer hosted Payments). a workshop in Dallas, TX, in which the conception and design of this Ricardo S. Avila project were developed. We appreciate the efforts of Pia Hirsch in the Employment: Accumetra, Paraxial, Kitware (I) coordination of that meeting. We appreciate David Gierada’s efforts in Leadership: Accumetra, Paraxial, Kitware (I) critically reviewing this manuscript. In addition, we thank Giorgio Stock and Other Ownership Interests: Accumetra, Paraxial, Kitware (I) Scagliotti, president of IASLC, and David Mesko, CEO of IASLC, for Consulting or Advisory Role: Cannot disclose (I), Cannot disclose ongoing support of this critical work. Patents, Royalties, Other Intellectual Property: Accumetra has patents pending Anand Devaraj Honoraria: Boehringer Ingelheim Consulting or Advisory Role: Boehringer Ingelheim, GlaxoSmithKline 98 © 2020 by American Society of Clinical Oncology IASLC-ELIC Initiative REFERENCES 1. Siegel RL, Miller KD, Jemal A: Cancer statistics, 2019. CA Cancer J Clin 69:7-34, 2019 2. Torre LA, Siegel RL, Jemal A: Lung cancer statistics, in Ahmad A, Gadgeel S (eds): Lung Cancer and Personalized Medicine. Advances in Experimental Medicine and Biology, Volume 893. Switzerland, Springer, Cham, 2016. DOI: https://doi.org/10.1007/978-3-319-24223-1_1 3. International Early Lung Cancer Action Program Investigators: Survival of patients with stage I lung cancer detected on CT screening. N Engl J Med 355:1763-1771, 2006 4. Aberle DR, Adams AM, Berg CD, et al: Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 365:395-409, 2011 5. De Koning H, Van Der Aalst C, Ten Haaf K, et al: Effects of volume CT lung cancer screening: Mortality results of the NELSON randomized-controlled population based trial. 2018 World Conference on Lung Cancer. Abstract PL02.05. Toronto, Ontario, Canada, September 25, 2018 6. Pastorino U, Silva M, Sestini S, et al: Prolonged lung cancer screening reduced 10-year mortality in the MILD trial: New confirmation of lung cancer screening efficacy. Ann Oncol 30:1162-1169, 2019 7. Nawa T, Fukui K, Nakayama T, et al: A population-based cohort study to evaluate the effectiveness of lung cancer screening using low-dose CT in Hitachi city, Japan. Jpn J Clin Oncol 49:130-136, 2019 8. Yankelevitz DF, Gupta R, Zhao B, et al: Small pulmonary nodules: Evaluation with repeat CT--preliminary experience. Radiology 212:561-566, 1999 9. van Klaveren RJ, Oudkerk M, Prokop M, et al: Management of lung nodules detected by volume CT scanning. N Engl J Med 361:2221-2229, 2009 10. Horeweg N, van Rosmalen J, Heuvelmans MA, et al: Lung cancer probability in patients with CT-detected pulmonary nodules: A prespecified analysis of data from the NELSON trial of low-dose CT screening. Lancet Oncol 15:1332-1341, 2014 11. Henschke CI, Yip R, Yankelevitz DF, et al: Definition of a positive test result in computed tomography screening for lung cancer: a cohort study. Ann Intern Med 158:246-252, 2013 12. Field JK, Duffy SW, Baldwin DR, et al: UK Lung Cancer RCT Pilot Screening Trial: Baseline findings from the screening arm provide evidence for the potential implementation of lung cancer screening. Thorax 71:161-170, 2016 13. Oudkerk M, Devaraj A, Vliegenthart R, et al: European position statement on lung cancer screening. Lancet Oncol 18:e754-e766, 2017 14. Rami-Porta R, Bolejack V, Crowley J, et al: The IASLC lung cancer staging project: Proposals for the revisions of the T descriptors in the forthcoming eighth edition of the TNM classification for lung cancer. J Thorac Oncol 10:990-1003, 2015 15. Lantuejoul S, Rouquette I, Brambilla E, et al: New WHO classification of lung adenocarcinoma and preneoplasia [in French]. Ann Pathol 36:5-14, 2016 16. Navale V, Bourne PE: Cloud computing applications for biomedical science: A perspective. PLOS Comput Biol 14:e1006144, 2018 17. Mulshine JL, Gierada DS, Armato SG III, et al: Role of the Quantitative Imaging Biomarker Alliance in optimizing CT for the evaluation of lung cancer screen- detected nodules. J Am Coll Radiol 12:390-395, 2015 18. Rydzak CE, Armato SG, Avila RS, et al: Quality assurance and quantitative imaging biomarkers in low dose CT lung cancer screening. Br J Radiol 91:20170401, 19. Field JK, deKoning H, Oudkerk M, et al: Implementation of lung cancer screening in Europe: Challenges and potential solutions: Summary of a multidisciplinary roundtable discussion. ESMO Open. 2019 Oct 13;4(5):e000577. doi: 10.1136/esmoopen-2019-000577. eCollection 2019 20. Lung Imaging Database Consortium: https://wiki.cancerimagingarchive.net/display/Public/LIDC-IDRI 21. IASLC News: Mulshine JL, Field JK: The IASLC’s spotlight on screening. http://www.lungcancernews.org/2019/02/26/the-iaslcs-spotlight-on-screening 22. Early Lung Imaging Confederation: http://iaslc-elic.org 23. ASCO Post: McNeil C: Low-dose CT lung screening: New developments support increased quality, more data, deep learning. https://www.ascopost.com/issues/ december-25-2018/low-dose-ct-lung-screening/ 24. US Food and Drug Administration: Proposed regulatory framework for modifications to artificial intelligence/machine learning (AI/ML)-based software as a medical device (SaMD). https://www.fda.gov/media/122535/download 25. Conley E, Pocs M: GDPR compliance challenges for interoperable health information exchanges (HIEs) and trustworthy research environments (TREs). Eur J Biomed Inform 14:48-61, 2018. https://www.ejbi.org/scholarly-articles/gdpr-compliance-challenges-for-interoperable-health-information-exchanges-hies- and-trustworthy-research-environments-tre.pdf 26. Mulshine JL: One screening for ischemic heart disease, lung cancer, and chronic obstructive pulmonary disease: A systems biology bridge for tobacco and radiation exposure. Am J Public Health 108:1294-1295, 2018 nn n JCO Clinical Cancer Informatics 99 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png JCO Clinical Cancer Informatics Wolters Kluwer Health

Loading next page...
 
/lp/wolters-kluwer-health/the-international-association-for-the-study-of-lung-cancer-early-lung-vg0cX0X4yf

References (58)

Publisher
Wolters Kluwer Health
Copyright
(C) 2020 American Society of Clinical Oncology
ISSN
2473-4276
DOI
10.1200/CCI.19.00099
Publisher site
See Article on Publisher Site

Abstract

review articles abstract The International Association for the Study of Lung Cancer Early Lung Imaging Confederation 1 2 3 4 5 James L. Mulshine, MD ; Ricardo S. Avila, MS ; Ed Conley, PhD ; Anand Devaraj, MD ; Laurie Fenton Ambrose, BS ; 6 7 8 9 10 Tanya Flanagan, MA ; Claudia I. Henschke, MD, PhD ; Fred R. Hirsch, MD, PhD ; Robert Janz, MSci ; Ryutaro Kakinuma, MD, PhD ; 11 12 13 13 14 Stephen Lam, MD ; Annette McWilliams, MBBS ; Peter M.A. Van Ooijen, PhD ; Matthijs Oudkerk, MD, PhD ; Ugo Pastorino, MD ; 15 16 16 17 13 18 Anthony Reeves, PhD ; Patrick Rogalla, MD ; Heidi Schmidt, MD ; Daniel C. Sullivan, MD ; Haije H.J. Wind, MSc ; Ning Wu, MD ; 19 20 7 3 Murry Wynes, PhD ; Xie Xueqian, MD, PhD ; David F. Yankelevitz, MD ; and John K. Field, PhD PURPOSE To improve outcomes for lung cancer through low-dose computed tomography (LDCT) early lung cancer detection. The International Association for the Study of Lung Cancer is developing the Early Lung Imaging Confederation (ELIC) to serve as an open-source, international, universally accessible environment to analyze large collections of quality-controlled LDCT images and associated biomedical data for research and routine screening care. METHODS ELIC is an international confederation that allows access to efficiently analyze large numbers of high- quality computed tomography (CT) images with associated de-identified clinical information without moving primary imaging/clinical or imaging data from its local or regional site of origin. Rather, ELIC uses a cloud-based infrastructure to distribute analysis tools to the local site of the stored imaging and clinical data, thereby allowing for research and quality studies to proceed in a vendor-neutral, collaborative environment. ELIC’s hub-and- spoke architecture will be deployed to permit analysis of CT images and associated data in a secure envi- ronment, without any requirement to reveal the data itself (ie, privacy protecting). Identifiable data remain under local control, so the resulting environment complies with national regulations and mitigates against privacy or data disclosure risk. RESULTS The goal of pilot experiments is to connect image collections of LDCT scans that can be accurately analyzed in a fashion to support a global network using methodologies that can be readily scaled to accrued databases of sufficient size to develop and validate robust quantitative imaging tools. CONCLUSION This initiative can rapidly accelerate improvements to the multidisciplinary management of early, curable lung cancer and other major thoracic diseases (eg, coronary artery disease and chronic obstructive pulmonary disease) visualized on a screening LDCT scan. The addition of a facile, quantitative CT scanner image quality conformance process is a unique step toward improving the reliability of clinical decision support with CT screening worldwide. JCO Clin Cancer Inform 4:89-99. © 2020 by American Society of Clinical Oncology Licensed under the Creative Commons Attribution 4.0 License INTRODUCTION tomography (CT) screening. The concept of de- veloping an early lung cancer image registry emerged Lung cancer is the most lethal cancer throughout the ASSOCIATED through a series of workshops sponsored by the In- CONTENT world, and it typically presents at a late stage when ternational Association for the Study of Lung Cancer 1,2 Data Supplement cure is unlikely. Recent reports have demonstrated (IASLC) as a critical opportunity to accelerate the pace Author affiliations the usefulness of low-dose computed tomography (CT) of innovation in improving the curative management and support screening in reducing lung cancer mortality in heavily information (if 3-7 for detection and intervention with early lung cancer. tobacco-exposed individuals. However, to achieve applicable) appear at the most efficient screening management, groups The IASLC mission is to improve lung cancer outcomes the end of this have incorporated a quantitative assessment of pul- through international and multidisciplinary collabora- article. monary nodule volume to guide the diagnostic tive efforts. This large collaborative image archival Accepted on 8-12 December 5, 2019 case–finding efforts within the screening process. and analysis effort builds on the established IASLC and published at In this fashion, the false-positive detection rate can be successes with the national and international lung ascopubs.org/journal/ 10-12 reduced from 28% to approximately 3%. This cancer IASLC TNM Staging Committee, which has cci on February 6, improves the screening cost and reduces the potential been undertaken in collaboration with the Union for 2020: DOI https://doi. for iatrogenic harm, which would be a critical aspect to International Cancer Control and American Joint org/10.1200/CCI.19. 00099 include in the global implementation of computed Committee on Cancer, as well as with the recently 89 Mulshine et al CONTEXT Key Objective Can an open-source cloud-based environment become a repository of screening computed tomography images and as- sociated data to enable quantitative and related imaging tool development for use in guiding the management of early, presymptomatic lung cancer and related thoracic diseases? Knowledge Generated A prototype international, open-source imaging resource is proposed that can allow federated image/data interrogation. This construct is designed to comply with existing international standards for data security while enabling the development and validation of new imaging biomarkers to facilitate early lung cancer management. Relevance Lung cancer screening is emerging as an important approach for early lung cancer management. However, robust and economic image analysis tools are needed to ensure facile clinical workflows for pulmonary nodule detection and quantitative nodule assessment so that this new service can be provided to the target high-risk population at high quality throughout the world. updated lung cancer pathology collaboration with the all of the images and metadata within their defined spoke 14,15 WHO. Success in developing screening tools, as with environment consistent with local governing data-sharing the staging and pathology efforts, involves issues of scale provisions. With local site permission, their stored clinical and cost that leverage the IASLC’s broad international and imaging data can be made accessible, to allow soft- scope and expertise in aligning global participation to ware tools distributed to the spoke from the hub to the improve early lung cancer management. relevant stored digital data. In this confederated architec- ture, the hub acts as a conduit with the spokes to dis- To date, no consensus image analysis tool has emerged to tribute software analysis tools to the relevant spokes and allow routine and reliable volumetric characterization of then to aggregate the results of the analysis of the images pulmonary nodules in routine clinical imaging settings. stored locally in the participating spokes. Therefore, only Developing a tool to allow easy and robust nodule mea- the resulting analysis data will leave an individual spoke. surement requires access to large numbers of high-quality The resulting analysis data aggregated in the hub from thoracic CT images that were acquired with the intent of multiple sites with diverse populations allow for research precisely measuring volumes of pulmonary nodules 5 to and for quality questions to be addressed with a poten- 10 mm in diameter. Because this is a newly appreciated tially vast number of test screening cases from multiple opportunity as lung imaging technology rapidly improves, countries. lung images from earlier CT screening trials were generally not acquired with sufficient resolution to enable this pre- To accommodate this architecture, the most cost-feasible cise volumetric tool development. Therefore, prospective approach for a flexible, scalable, and sustainable envi- collections of CT images from current-generation, high- ronment capable of enabling the goals of ELIC is to leverage resolution CT scanners are urgently required to address the global accessibility of a cloud environment. Currently, thoracic CT screening images and associated clinical this gap. This lack of large quantities of such high-quality outcomes and relevant metadata are stored in a vast array image data imposes a profound barrier to progress with of architectures across IASLC member sites. We envision early lung cancer management. developing a vendor-neutral, secure, scalable, cloud-based How to Bring Value to Early Lung Cancer Detection environment to bridge to existing sites’ data storage re- In response to this situation, the IASLC hosted a planning sources. Table 1 summarizes the design considerations workshop held in Dallas, Texas, in February 2018. Al- guiding the development of this informatics resource. Given though aware of the heterogeneous nature of existing image the dynamic and complex nature of the privacy challenges registries at leading centers from around the world, the inherent in collecting and sharing large amounts of imaging group proposed the creation of a cloud-based informatics and clinical data, the proposed IASLC imaging/data re- infrastructure to interact with existing international regis- source may be preferable for many national sponsors to tries and centers collecting thoracic CT images together have a rigorously designed, precompetitive environment with associated core clinical outcomes data to optimize cost hosted by an international, nonprofit professional society and data security. such as IASLC. IASLC has a proven legacy of patient This IASLC Early Lung Imaging Confederation (ELIC) was benefit and as a reliable host to ensure appropriate proposed as a hub-and-spoke architecture with the in- stewardship as an “honest broker” for such a critical in- 14,15 tention of enabling the imaging-donating local site to retain ternational resource. 90 © 2020 by American Society of Clinical Oncology IASLC-ELIC Initiative TABLE 1. Guiding Principles of ELIC acquisition protocols for CT lung nodule measurement. The data are also revealing new insights into low-dose CT lung In alignment with its mission and charge, ELIC seeks to share its resource with both cancer screening image acquisition performance, in- national and international scientific lung cancer research communities by: cluding the reproducibility of image acquisition properties, Promoting the common good by maximizing collaborative research for the differences between requested and obtained slice thick- benefit of all ness, and losses of resolution associated with lowering the Making ELIC research data available to the research community to advance radiation dose. scientific knowledge Providing open-source resources including an environment and tools to METHODS ensure broadest dissemination of resources to enhance outcomes with A proof-of-concept demonstration was developed in 2018 tobacco-related diseases to evaluate the potential of a hub-and-spoke–distributed Ensuring the generation of high-quality research lung cancer screening image archive and computing ar- Respecting the legal rights and legitimate interests of all stakeholders involved chitecture to achieve the goals of ELIC . Ten international Protecting the privacy of its research participants and the confidentiality of cloud computing sites were chosen from the Amazon Web their data Services global cloud network map to launch and set-up Promoting transparency, responsibility, interoperability, and fairness spoke EC2 cloud computing instances, as shown in Figure Ensuring accountability and oversight 1. This figure overlays the 10 ELIC spokes feeding a central hub (at the point of the arrow) on the Amazon Web Services Enriching the content of the ELIC database, including through the return of high quality–derived data by researchers (AWS) global cloud network map. The hub is shown (with the blue circle at the tip of the arrow), with each of the 10 spokes Managing access to ELIC data to balance current and future needs (indicated by green circles) populated with an identical set of Abbreviation: ELIC, Early Lung Imaging Confederation. 100 publicly available de-identified CT lung scans. As shown on the map, existing AWS cloud services sites are indicated by smaller (lighter) blue and purple circles. This distribution High-Quality Image Acquisition of existing cloud resources demonstrate the global reach of For optimal efficiency, there is also the need to pro- AWS to support local/regional hosting of available lung spectively ensure that newly accrued thoracic CT images cancer screening images and corresponding clinical data as are objectively of sufficiently high quality to support robust required by General Data Protection Regulations. Each of quantitative analysis. Because ELIC is envisioned to be these spokes was populated with an identical set of 100 a resource to develop and then validate image processing publicly available de-identified CT lung scans. However, algorithms as required to support clinical decision making each scan was given a unique patient ID and fictitious age, for early lung cancer management, such efforts will move sex, and pack-year demographics. This was done so that into the rapidly evolving realm of deep learning methods or proof-of-concept performance testing could be obtained with artificial intelligence approaches; therefore, the curation of the equivalent of 1,000 patient cases; the analyses de- an image collection that has been optimized for quantitative scribed in this report were repeated on at least 5 separate 17,18 measurement is critical. occasions. In addition, a Hub EC2 cloud computing instance Reliable guidance and clinical management in the lung was set up at the Northern Virginia location of the Amazon cancer screening setting requires accurate and robust Web Services cloud. analysis of pulmonary nodule volume. Therefore, ELIC will Figure 2 is a schematic representation of the computing require image-sharing sites to prospectively adhere to the resources and main data flows that were used during the Quantitative Imaging Biomarkers Alliance (QIBA) of the design of the ELIC Hub-and-Spoke Environment (H&SE) Radiologic Society of North America (RSNA) CT Small Lung software. A lung cancer screening principal investigator Nodule Profile quality provisions to ensure that aggregated (PI; A) typically coordinates CT lung cancer screening imaging data obtained for quantitative analysis are of a high for several screening sites shown as sources (Ai). The PI 19,20 quality; this is a unique strength of the ELIC effort. is responsible for submitting de-identified lung cancer From extensive work in optimizing image measurement screening image DICOM (Digital Imaging and Communi- quality in this setting, the QIBA has developed a process to cations in Medicine) data and metadata, including any 17,18 address these factors. This entails analyzing pulmonary requested image annotations, before data upload to a nodules in the volume range relevant to lung cancer spoke (B). Each time data on the spoke (B) are added or screening (6-10 mm in diameter). Crowd-sourced data corrected, a set of basic demographics and metadata in- collected from international QIBA CT Small Lung Nodule formation about the new data is communicated to the hub testing sites have been used over the past 2 years to help (C). In this way, the hub and spoke will remain in agreement identify and globally disseminate the top-performing CT regarding the data that a PI has made available for analysis image acquisition protocols for each CT scanner model. on the spoke (B). Because the spoke (B) can be set up on This innovative approach has enabled what we believe to a local cloud computing instance or on local computing be the first rapid global optimization of scanner image hardware, the spoke (B) data will remain within the source JCO Clinical Cancer Informatics 91 Mulshine et al Phantom image quality testing Edge locations Multiple edge locations Regional edge caches FIG 1. The global distribution of the hub and spokes for the 2018 World Conference on Lung Cancer ELIC proof-of-concept demonstrations using the Amazon Web Services (AWS) cloud. geographic region and will only be analyzed according to contain fully automated scripts for thoroughly performing strictly administered analysis and reporting rules (D) de- required data protection operations, such as the General cided by the PI (A). The data residing on all spokes are de- Data Protection Regulation “right to be forgotten.” Each identified by the PI (A) before upload. ELIC spoke will have automated scripts that achieve the data governance protections and procedures required by Thus, at all times, only de-identified data are stored on the local regulations, some of which will be standard and can ELIC H&SE and, at rest, the data are also encrypted. Each be provided by the same ELIC H&SE script for all regions. ELIC spoke contains a de-identified set of data stored on a spoke using identical ELIC H&SE data storage organi- A client (E) can view a listing or catalog of all the data sets zation and methods. This allows the ELIC H&SE software to available within the ELIC environment and take the Source ELIC Environment Screening Source A PI Source Client Cloud Basic Meta Instance Hub Spoke Algor gorithm Cloud Instance Res esul u ts FIG 2. High-level illustration of the ELIC Hub & Spoke Environment and how lung cancer screening principal in- vestigators (PI) and clients will interact with the system. DICOM, Digital Imaging and Communications in Medicine; ELIC, Early Lung Imaging Confederation. 92 © 2020 by American Society of Clinical Oncology Results Algorithm Basic Meta DICOM Meta Data IASLC-ELIC Initiative necessary steps to run a quantitative algorithm on all the the quality of a CT scanner and acquisition process. The data or on a subset of the data available within the ELIC results of the comprehensive QIBA CT image quality environment. This is referred to as running an “experiment” analysis are sent back to the site within a few minutes using within the ELIC environment. To run an experiment, the an easy-to-interpret structured report indicating whether client (E) must provide a quantitative algorithm according to the CT scanner and acquisition parameters were of suffi- ELIC H&SE specifications to the hub (C) and request that cient quality for quantitative assessment. If the image the hub execute the experiment on all the relevant spokes quality was insufficient, then remediation measures are (B). The hub (C) then coordinates the running of the suggested to the site to improve image quality. This analysis quantitative algorithm on the spokes and collects the re- has been performed at. 60 CT lung cancer screening sites sults according to the analysis rules (D) agreed to by the PI. evaluating the use of the CTLX1 phantom, including When the experiment is finished, the client (E) is provided screening sites in Australia, England, Canada, China, Israel, an aggregate summary of all the data collected from the Italy, the Netherlands, Poland, Spain, Japan, and the experiment, including information on any errors or issues United States. The image quality data collected from the encountered while running the experiment. A prototype analysis of CTLX1 phantom scans include CT scanner hub password-protected Website was created that dis- image acquisition parameter settings such as milliampere, plays the status of spokes, allows the specification and peak kilovoltage, slice thickness, slice spacing, and re- construction kernel, as well as fundamental image quality launching of quantitative lung imaging experiments on characteristics achieved, including levels of edge en- global collections of data, and provides a real-time quan- hancement, 3-dimensional (3D) resolution, 3D resolution titative and graphic display of the results obtained from the aspect ratio, CT linearity, noise, and 3D spatial warping. international spoke calculations. Each of these fundamental image quality properties is Two cloud-based experiments using software algorithms measured throughout the CT scanner field of view at 3 were created to accept DICOM data and produce quanti- distances from isocenter (0, 100, and 200 mm) to ensure tative results and images in a standard format that the ELIC that lung nodules present in the lung periphery, which is H&SE supports. One algorithm quantitatively measured common, can be accurately measured. total lung volume, and the second algorithm quantitatively For proof-of-concept testing, the QIBA CT Small Lung measured the volume of solid pulmonary nodules. This Nodule Profile Conformance automated phantom analysis permitted the ELIC H&SE to demonstrate running the software was placed on SPOKE 1 running in northern quantitative imaging experiments on image collections Virginia and used to run automated image quality analyses at globally distributed spokes and then aggregating the on scans of the QIBA CT Small Lung Nodule Profile CTLX1 quantitative CT image measurements and output image phantom. These tests confirmed that the fully automated results on the hub. Each of the 10 spokes was set up to run QIBA CT Small Lung Nodule Profile conformance certifi- one of the 2 quantitative CT lung image measurement cation methods for CT image quality assessment will be algorithms when requested by the hub. able to successfully run on future ELIC spokes (or hub) The project also deployed the QIBA CT Small Lung Nodule running on the Amazon cloud. Profile Conformance Certification service on the hub and A total of 5 live demonstrations were run using the ELIC performed CT image quality conformance assessment environment at 5 distinct time windows to evaluate the calculations on CT phantom scans. Cloud-based phantom ability of ELIC H&SE to perform useful quantitative imaging analysis software was developed to perform a low-cost CT computational experiments on large collections of globally image quality assessment using a specifically designed distributed CT lung cancer imaging cases without moving phantom (ie, test object). This approach makes achieving data out of geographic regions. Each demonstration was CT image quality conformance with the QIBA Small Lung run for ELIC H&SE analyses performed with a central hub Nodule Profile possible from virtually any clinical imaging 17,18 server and 10 globally distributed spoke servers all running site in the world. The purpose of this quality control step on the Amazon Web Services cloud. The prototype ELIC H& is to ensure consistent image quality appropriate for the SE Website allowed a user to launch a computational defined context of use with lung cancer screening–related experiment request on a specific set of data distributed quantitative assessment for imaging sites around the world. across any number of spokes, store the results, and display The standardized assessment of CT lung imaging quality is an aggregated summary report of the results when the enabled using a dedicated, low-cost phantom (CTLX1) that analysis was complete. was developed for this purpose. The CTLX1 phantom contains small, precision-made, geometric components to In all, 3 pilot testing runs were performed. For these pilot assess thoracic CT imaging performance. A CT scan of the testing runs, quantitative lung volume analyses were CTLX1 phantom is typically acquired in approximately completed on 1,000 globally distributed CT lung cancer 5 minutes and then uploaded to the QIBA Phantom screening data sets in , 25 minutes at 10 different ELIC Analysis Service, which rapidly analyzes the uploaded hosting sites from 4 continents. Figure 3 shows the main phantom image using automated software to characterize user page available within the Website, which includes JCO Clinical Cancer Informatics 93 Mulshine et al FIG 3. The main user page available at www.iaslc-elic.org for the launch and review of quantitative lung imaging experiments on globally distributed spokes. a listing of the available spokes, a place to specify the software is set up so that each spoke performs a compu- launch of a quantitative experiment on the data located on tational request and returns quantitative results back to the the spokes, and a list of pages showing the results obtained hub. Figure 4 shows the results page from a representative from each experiment. lung nodule measurement experiment (experiment 3) performed on all the images at the 10 sites in the archive The developed H&SE software allows a hub server to make with lung nodules (N = 62 × 10) in the ELIC H&SE. The quantitative CT imaging computational requests to a col- results reported were mean volume and standard deviation lection of globally distributed spokes, each of which is populated with de-identified CT lung images. The H&SE for all the data sets analyzed. Figure 5 displays the detailed 94 © 2020 by American Society of Clinical Oncology IASLC-ELIC Initiative FIG 4. Results page for a lung nodule measurement experiment performed in the ELIC Hub and Spoke Experiment. quantitative imaging results that were computed on CT lung Toronto, Canada. In total, . 5 live demonstrations suc- imaging cases on SPOKE 10 (Sao Paulo, Brazil) for the cessfully showed the simultaneous running of the CT lung nodule algorithm on CT lung imaging cases distributed at same experiment as that shown in Figure 4 (experiment 3), 10 internationally distributed spoke sites. including the quantitative lung nodule volume values and 3 orthogonal reformat images with contours of the segmen- DISCUSSION tation results overlaid for each case. This initial proof-of-concept project demonstrated the po- Three presentations, including a live demonstration, were tential of the ELIC H&SE to provide a useful resource for given at IASLC’s 2018 World Conference on Lung Cancer in global quantitative lung imaging studies. This project used JCO Clinical Cancer Informatics 95 Mulshine et al FIG 5. Detailed results for cases run on spoke 10 (Sao Paulo) for experiment 13. global cloud-computing resources, each populated with an QIBA CT Small Lung Nodule Profile CT image quality identical data set of 100 publicly available lung cancer conformance testing service was successfully run on a spoke to demonstrate that CT image quality monitoring images. A central hub Website simultaneously distributed and optimization can be supported with the ELIC H&SE. 2 open-source quantitative lung measurement algorithm requests to the 10 globally distributed spoke sites. Using Although the developed proof-of-concept software imple- publicly available CT lung images allowed the project to mented a basic H&SE, the functionality represents a subset achieve results quickly. The hub received and aggregated of that which will be needed to fully realize the IASLC ELIC all quantitative algorithm results from the spoke sites and H&SE vision. For example, the demonstration Website displayed the results on the hub Website. In addition, the supports only 1 user type; when mature, the ELIC H&SE 96 © 2020 by American Society of Clinical Oncology IASLC-ELIC Initiative will support a variety of different roles and types of access of the processes for the users of the spokes and hub re- for clients, hub personnel, and spoke personnel. Future quires a well-functioning informatics environment with versions of the ELIC H&SE will also provide higher levels of easy-to-setup and deployment tools that will enable rapid security to prevent access or extraction of PI CT image data screening care implementation as well as research by new as well as to prevent unauthorized use of client algorithms. global lung cancer screening groups. For these initial demonstrations, the client algorithms were The power of ELIC relates to the accessibility of the cloud created as open-source projects and built on each spoke and its remarkable cost efficiency. For image quality into an executable application. However, a future ELIC H& processes, ELIC already uses machine vision and will soon SE will likely also support the distribution of algorithms from include artificial intelligence to ensure optimal and eco- the hub. For example, the distribution of virtual containers nomic image quality. This resource can greatly accelerate with executable applications embedded is being explored. radiomics and deep learning processes for medical images Future H&SE environmentswillalsoneedtoprovideadditional and can be integrated with digital pathology and genomic functionality andsupportforartificialintelligencedevelopment data. As the ELIC project continues to mature as a resource and specifically, deep learning algorithm development and to conduct analyses and study international collections of testing. A fuller description of the methodology, governance, high-quality thoracic CT images together with associated data use provisions, scope, and other proposed policy ele- biomedical data, there are a number of models through ments are provided in the Data Supplement. which the pharmaceutical industry can be involved. For The ELIC H&SE prototype vehicle was developed with example, drug company sponsors may host their own ELIC IASLC seed funding to support feasibility testing and was spoke, on which they can collect quality-controlled CT developed as an open-science and open-source research/ images associated with new innovative neoadjuvant stud- care environment to allow a broad range of collaborative ies, adjuvant or even chemoprevention clinical trials, and participation in achieving its CT imaging-related objectives. associated clinical outcomes data. With each trial, the Funding to sustain this large-scale international collabo- sponsor can decide which image collections to acquire and rative effort will be dependent on the early lung cancer maintain as private versus aggregated in large public image research and innovation communities using ELIC in their collections. In this fashion, aggregated image and data research to address important research and lung cancer collections could emerge as critical postmarketing research care issues. We expect many stakeholders, including and quality resources. Furthermore, access to large quality- software developers, artificial intelligence companies, im- controlled CT screening images with clinical outcomes data aging researchers, diagnostic device developers, medical will provide the basis on which to construct the next imaging companies, government agencies, pharmaceuti- generation of algorithms to build models and tools, which cal developers, academic societies, and many other in- can function as clinical decision support. This is a rapidly novators, to be early users of the flexible capabilities of evolving area, and we are working to evolve ELIC in ELIC. This service model is designed to allow a sustainable alignment with new regulatory guidance relative to re- path forward for this internationally accessible quantitative sponsibly building international clinical decision support 23-25 imaging environment to emerge as a core resource for tools for clinical care and research. Because thoracic improving the detection and management of early thoracic CT images from screening also contain information about disease. the presence of early coronary artery disease and chronic The internationally federated ELIC hub-and-spoke system obstructive disease, a high-quality international collection of these images will be of growing public health value. has features such as robust data privacy provisions to permit large-scale analyses of clinical CT images with In summary, the development and deployment of the ELIC relevant associated data in a secure environment (ie, pri- hub-and-spoke environment, together with fair and inter- vacy protection that is under the control of the Trial/Study nationally developed governance policies, will establish, to PI), so that this resource could support the conduct of our knowledge for the first time, a large and efficient global clinical trials. This approach is intended to ensure local computing environment for the study of thoracic CT scans governance control of the site PI, who can address the obtained in the context of lung cancer screening. The use of specific data protection conditions at diverse international QIBA CT Small Lung Nodule Profile and the associated locations. phantoms and software tools will improve the quality of In addition to prospective accumulation of individual new global thoracic CT images aggregated for ELIC, ensuring screening cases, existing imaging collections will remain at much more efficient image tool development. Not only will the local site where they were collected, so, again, the these resources help accelerate lung imaging research and resulting environment remains consistent with local regu- the availability of thoroughly tested imaging tools, but the lations without a privacy or data disclosure risk. An inherent data generated will provide insights to guide future rec- design feature of ELIC is to manage large numbers of ommendations for lung screening and for managing early thoracic CT image screening registries. Therefore, control thoracic diseases. JCO Clinical Cancer Informatics 97 Mulshine et al Fred R. Hirsch AFFILIATIONS Consulting or Advisory Role: AstraZeneca, Genentech, Lilly, Merck, Rush University, Chicago, IL Bristol-Myers Squibb, Pfizer, Roche, Loxo, Novartis, Amgen, Oncocyte Accumetra, LLC, Clifton Park, NY Research Funding: Amgen (Inst), Merck (Inst), Abbvie (Inst), Biodesix University of Liverpool, Liverpool, United Kingdom (Inst), Mersana (Inst), Rain Therapeutics (Inst) University College, London, United Kingdom Patents, Royalties, Other Intellectual Property: EGFR FISH and IHC for GO2 Foundation, Washington, DC prediction of outcome in patients treated with EGFR inhibitors (Inst) Canadian Partnership Against Cancer, Toronto, Ontario, Canada Icahn School of Medicine at Mount Sinai, New York, NY Stephen Lam Mount Sinai Health System, New York, NY Patents, Royalties, Other Intellectual Property: Deep learning prediction University of Groningen, Groningen, Netherlands algorithm to estimate the 3-year lung cancer risk and cancer related National Cancer Center Hospital, Tokyo, Japan mortality for individuals who have . 2 screening chest CT scans. Joint University of British Columbia, Vancouver, British Columbia, Canada application by Johns Hopkins University and the BC Cancer Agency Fiona Stanley Hospital, Western Australia Patent pending (Inst). University Medical College, Groningen, Netherlands Instituti Tumori, Milan, Italy Annette McWilliams Cornell University, Ithaca, New York, NY Travel, Accommodations, Expenses: Olympus Medical Systems, Roche Toronto Joint Department of Medical Imaging, University of Toronto, Matthijs Oudkerk Ontario, Canada Employment: Institute for Diagnostic Accuracy Duke University Medical Center, Durham, NC Speakers’ Bureau: Siemens Healthineers National Cancer Center, Peking Union Medical College, Beijing, China Research Funding: EU Horizon 2020 (Inst) International Association for the Study of Lung Cancer, Denver, CO 20 Travel, Accommodations, Expenses: AstraZeneca Shanghai General Hospital, Shanghai, China Anthony Reeves CORRESPONDING AUTHOR Leadership: D4Vision James L. Mulshine, MD, Rush University Medical Center, 1700 W Van Stock and Other Ownership Interests: VisionGate Buren St, Suite 245, Triangle Office Building, Chicago, IL 60612; Patents, Royalties, Other Intellectual Property: Co-inventor on patents e-mail: jmulshin@rush.edu. owned by Cornell Research Foundation, which are nonexclusively licensed and are related to technology involving computer-aided diagnostic methods. AUTHOR CONTRIBUTIONS Conception and design: All authors Patrick Rogalla Collection and assembly of data: Ricardo S. Avila Research Funding: Canon Medical (Inst) Data analysis and interpretation: James Mulshine, Ricardo S. Avila, John Travel, Accommodations, Expenses: Canon Medical (Inst) Field David F. Yankelevitz Manuscript writing: All authors Stock and Other Ownership Interests: Accumetra Final approval of manuscript: All authors Consulting or Advisory Role: Grail Accountable for all aspects of the work: All authors Patents, Royalties, Other Intellectual Property: Licensing agreement between Cornell University and General Electric for management of AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF various lung abnormalities including lung nodules. INTEREST John K. Field The following represents disclosure information provided by authors of Consulting or Advisory Role: Epigenomics (Inst), NUCLEIX (Inst), this manuscript. All relationships are considered compensated unless AstraZenca (Inst), AstraZenca (Inst), Institute Diagnostic Accuracy (Inst) otherwise noted. Relationships are self-held unless noted. I = Immediate Speakers’ Bureau: AstraZenca (Inst) Family Member, Inst = My Institution. Relationships may not relate to the Research Funding: Janssen Research & Development (Inst) subject matter of this manuscript. For more information about ASCO’s conflict of interest policy, please refer to www.asco.org/rwc or ascopubs. No other potential conflicts of interest were reported. org/cci/author-center. Open Payments is a public database containing information reported by ACKNOWLEDGMENT companies about payments made to US-licensed physicians (Open The International Association for the Study of Lung Cancer hosted Payments). a workshop in Dallas, TX, in which the conception and design of this Ricardo S. Avila project were developed. We appreciate the efforts of Pia Hirsch in the Employment: Accumetra, Paraxial, Kitware (I) coordination of that meeting. We appreciate David Gierada’s efforts in Leadership: Accumetra, Paraxial, Kitware (I) critically reviewing this manuscript. In addition, we thank Giorgio Stock and Other Ownership Interests: Accumetra, Paraxial, Kitware (I) Scagliotti, president of IASLC, and David Mesko, CEO of IASLC, for Consulting or Advisory Role: Cannot disclose (I), Cannot disclose ongoing support of this critical work. Patents, Royalties, Other Intellectual Property: Accumetra has patents pending Anand Devaraj Honoraria: Boehringer Ingelheim Consulting or Advisory Role: Boehringer Ingelheim, GlaxoSmithKline 98 © 2020 by American Society of Clinical Oncology IASLC-ELIC Initiative REFERENCES 1. Siegel RL, Miller KD, Jemal A: Cancer statistics, 2019. CA Cancer J Clin 69:7-34, 2019 2. Torre LA, Siegel RL, Jemal A: Lung cancer statistics, in Ahmad A, Gadgeel S (eds): Lung Cancer and Personalized Medicine. Advances in Experimental Medicine and Biology, Volume 893. Switzerland, Springer, Cham, 2016. DOI: https://doi.org/10.1007/978-3-319-24223-1_1 3. International Early Lung Cancer Action Program Investigators: Survival of patients with stage I lung cancer detected on CT screening. N Engl J Med 355:1763-1771, 2006 4. Aberle DR, Adams AM, Berg CD, et al: Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 365:395-409, 2011 5. De Koning H, Van Der Aalst C, Ten Haaf K, et al: Effects of volume CT lung cancer screening: Mortality results of the NELSON randomized-controlled population based trial. 2018 World Conference on Lung Cancer. Abstract PL02.05. Toronto, Ontario, Canada, September 25, 2018 6. Pastorino U, Silva M, Sestini S, et al: Prolonged lung cancer screening reduced 10-year mortality in the MILD trial: New confirmation of lung cancer screening efficacy. Ann Oncol 30:1162-1169, 2019 7. Nawa T, Fukui K, Nakayama T, et al: A population-based cohort study to evaluate the effectiveness of lung cancer screening using low-dose CT in Hitachi city, Japan. Jpn J Clin Oncol 49:130-136, 2019 8. Yankelevitz DF, Gupta R, Zhao B, et al: Small pulmonary nodules: Evaluation with repeat CT--preliminary experience. Radiology 212:561-566, 1999 9. van Klaveren RJ, Oudkerk M, Prokop M, et al: Management of lung nodules detected by volume CT scanning. N Engl J Med 361:2221-2229, 2009 10. Horeweg N, van Rosmalen J, Heuvelmans MA, et al: Lung cancer probability in patients with CT-detected pulmonary nodules: A prespecified analysis of data from the NELSON trial of low-dose CT screening. Lancet Oncol 15:1332-1341, 2014 11. Henschke CI, Yip R, Yankelevitz DF, et al: Definition of a positive test result in computed tomography screening for lung cancer: a cohort study. Ann Intern Med 158:246-252, 2013 12. Field JK, Duffy SW, Baldwin DR, et al: UK Lung Cancer RCT Pilot Screening Trial: Baseline findings from the screening arm provide evidence for the potential implementation of lung cancer screening. Thorax 71:161-170, 2016 13. Oudkerk M, Devaraj A, Vliegenthart R, et al: European position statement on lung cancer screening. Lancet Oncol 18:e754-e766, 2017 14. Rami-Porta R, Bolejack V, Crowley J, et al: The IASLC lung cancer staging project: Proposals for the revisions of the T descriptors in the forthcoming eighth edition of the TNM classification for lung cancer. J Thorac Oncol 10:990-1003, 2015 15. Lantuejoul S, Rouquette I, Brambilla E, et al: New WHO classification of lung adenocarcinoma and preneoplasia [in French]. Ann Pathol 36:5-14, 2016 16. Navale V, Bourne PE: Cloud computing applications for biomedical science: A perspective. PLOS Comput Biol 14:e1006144, 2018 17. Mulshine JL, Gierada DS, Armato SG III, et al: Role of the Quantitative Imaging Biomarker Alliance in optimizing CT for the evaluation of lung cancer screen- detected nodules. J Am Coll Radiol 12:390-395, 2015 18. Rydzak CE, Armato SG, Avila RS, et al: Quality assurance and quantitative imaging biomarkers in low dose CT lung cancer screening. Br J Radiol 91:20170401, 19. Field JK, deKoning H, Oudkerk M, et al: Implementation of lung cancer screening in Europe: Challenges and potential solutions: Summary of a multidisciplinary roundtable discussion. ESMO Open. 2019 Oct 13;4(5):e000577. doi: 10.1136/esmoopen-2019-000577. eCollection 2019 20. Lung Imaging Database Consortium: https://wiki.cancerimagingarchive.net/display/Public/LIDC-IDRI 21. IASLC News: Mulshine JL, Field JK: The IASLC’s spotlight on screening. http://www.lungcancernews.org/2019/02/26/the-iaslcs-spotlight-on-screening 22. Early Lung Imaging Confederation: http://iaslc-elic.org 23. ASCO Post: McNeil C: Low-dose CT lung screening: New developments support increased quality, more data, deep learning. https://www.ascopost.com/issues/ december-25-2018/low-dose-ct-lung-screening/ 24. US Food and Drug Administration: Proposed regulatory framework for modifications to artificial intelligence/machine learning (AI/ML)-based software as a medical device (SaMD). https://www.fda.gov/media/122535/download 25. Conley E, Pocs M: GDPR compliance challenges for interoperable health information exchanges (HIEs) and trustworthy research environments (TREs). Eur J Biomed Inform 14:48-61, 2018. https://www.ejbi.org/scholarly-articles/gdpr-compliance-challenges-for-interoperable-health-information-exchanges-hies- and-trustworthy-research-environments-tre.pdf 26. Mulshine JL: One screening for ischemic heart disease, lung cancer, and chronic obstructive pulmonary disease: A systems biology bridge for tobacco and radiation exposure. Am J Public Health 108:1294-1295, 2018 nn n JCO Clinical Cancer Informatics 99

Journal

JCO Clinical Cancer InformaticsWolters Kluwer Health

Published: Feb 6, 2020

There are no references for this article.