Periodic Reporting for period 2 - BIGPICTURE (Central Repository for Digital Pathology)
Okres sprawozdawczy: 2022-08-01 do 2024-01-31
Artificial intelligence (AI) has the potential to enhance speed, precision, quality, and effectiveness of numerous activities across medical fields, also in pathology. Traditionally, pathologists diagnose and understand disease by examining tissues under a microscope or using digitised microscopy images. To leverage AI in pathology, AI algorithms are developed using substantial number and variety of images. Bigpicture's mission is to collect and share an unprecedented amount of quality-controlled pathology images and metadata, as well as AI algorithms, to facilitate research and development.
Building on existing assets such as derived from ELIXIR infrastructure, we established the first European ethical and regulatory compliant platform connecting pathologists, researchers, AI developers, patients, and industry. Our vision is to become a catalyst in the digital transformation of pathology, allowing AI to reach maturity. Bigpicture will enable the development and validation of trustworthy technologies to help diagnose and predict diseases and improve the quality and efficiency of drug development. We will create the tools and workflows to support the collection and sharing of large series of images and metadata by European pathology departments. By engaging and building stakeholder consensus, Bigpicture will contribute to a regulatory framework for digital pathology and AI-based methods. Finally, Bigpicture envisions sustainability of its platform through a community-based model that creates value and reciprocity in use.
Building on existing assets such as derived from ELIXIR infrastructure, we established the first European ethical and regulatory compliant platform connecting pathologists, researchers, AI developers, patients, and industry. Our vision is to become a catalyst in the digital transformation of pathology, allowing AI to reach maturity. Bigpicture will enable the development and validation of trustworthy technologies to help diagnose and predict diseases and improve the quality and efficiency of drug development. We will create the tools and workflows to support the collection and sharing of large series of images and metadata by European pathology departments. By engaging and building stakeholder consensus, Bigpicture will contribute to a regulatory framework for digital pathology and AI-based methods. Finally, Bigpicture envisions sustainability of its platform through a community-based model that creates value and reciprocity in use.
A management and governance structure was implemented and a Bigpicture community has been established with regular project meetings, newsletters, webinars, website, and social media. All partners are highly engaged in the project and a spirit of active collaboration between the work packages has been achieved, demonstrated by the active participation at numerous consortium meetings, and Advisory boards established.
The backend technical infrastructure has moved into production mode and data submissions have been achieved through large scale data submission API interfaces. Cybersecurity penetration testing was performed. Data mirroring between international data centers has been put in place. Discoverability services have been developed including public facing landing pages for datasets.
To support the collection of harmonized and valuable data, we developed a (meta)data model for all directly accessible datasets. The node coordinator network has started dataset preparations and 4 clinical partners have submitted data. 480.000 clinical WSI are in preparation for submission, of which 260.000 are ready for upload. The non-clinical data transporter has been deployed at multiple EFPIA sites and data has been transferred to the repository by 2 parties. As part of the honest broker mechanism the REMS service to manage data access requests and Perun group management services have been implemented and are integrated with LS-AAI.
To improve the usability of the platform, tools for conversion and ingestion of Whole Slide Images (WSI) DICOM and metadata, including tutorials and example implementations have been delivered. Tools for opening, viewing, searching, and annotating images and metadata have been developed. A first prototype implementation of algorithms for AI-based annotation, WSI registration and content-based object retrieval was established, and 4 publications on AI development have been published. A generic artifact detection algorithm was developed and validated with Bigpicture data. Additionally, we developed a weakly-supervised learning methodology, that allows end-to-end training with WSI and slide- or patient-level annotations. Lastly the technical description of container-based infrastructure to run AI algorithms, AI algorithm template, and first AI methodological repository has been made available.
The basics for responsible data sharing namely the DPIA, pseudonymisation strategy and the ethics advisory board have been delivered. A comparison of applicable national ethical & legal regulations and practices in relevant countries was completed. A key achievement is the Data Sharing Agreement, regulating data sharing between Bigpicture partners is now nearly ready for sign-off.
To plan for a sustainable future of the platform a market and stakeholder analysis was performed, and different organisational and governance models have been evaluated that will support the sustainable platform. Additionally, interactions with different stakeholders including patients have served as a basis for discussions on the acceptable business models for the platform.
The backend technical infrastructure has moved into production mode and data submissions have been achieved through large scale data submission API interfaces. Cybersecurity penetration testing was performed. Data mirroring between international data centers has been put in place. Discoverability services have been developed including public facing landing pages for datasets.
To support the collection of harmonized and valuable data, we developed a (meta)data model for all directly accessible datasets. The node coordinator network has started dataset preparations and 4 clinical partners have submitted data. 480.000 clinical WSI are in preparation for submission, of which 260.000 are ready for upload. The non-clinical data transporter has been deployed at multiple EFPIA sites and data has been transferred to the repository by 2 parties. As part of the honest broker mechanism the REMS service to manage data access requests and Perun group management services have been implemented and are integrated with LS-AAI.
To improve the usability of the platform, tools for conversion and ingestion of Whole Slide Images (WSI) DICOM and metadata, including tutorials and example implementations have been delivered. Tools for opening, viewing, searching, and annotating images and metadata have been developed. A first prototype implementation of algorithms for AI-based annotation, WSI registration and content-based object retrieval was established, and 4 publications on AI development have been published. A generic artifact detection algorithm was developed and validated with Bigpicture data. Additionally, we developed a weakly-supervised learning methodology, that allows end-to-end training with WSI and slide- or patient-level annotations. Lastly the technical description of container-based infrastructure to run AI algorithms, AI algorithm template, and first AI methodological repository has been made available.
The basics for responsible data sharing namely the DPIA, pseudonymisation strategy and the ethics advisory board have been delivered. A comparison of applicable national ethical & legal regulations and practices in relevant countries was completed. A key achievement is the Data Sharing Agreement, regulating data sharing between Bigpicture partners is now nearly ready for sign-off.
To plan for a sustainable future of the platform a market and stakeholder analysis was performed, and different organisational and governance models have been evaluated that will support the sustainable platform. Additionally, interactions with different stakeholders including patients have served as a basis for discussions on the acceptable business models for the platform.
Setting up a resource for AI development in pathology of the size of Bigpicture is unprecedented, attracting international attention from researchers, large and small companies. The numbers of partners, disease areas, and the unique combination of data from the clinical and preclinical field offer new perspectives on the use of AI in pathology. Bigpicture is on track, meeting project timelines and exceeding expectations beyond the state-of-the-art in a least four areas, addressed below.
Progress in this field is limited by a lack of standards, severely limiting interoperability and integration of AI-tools in the pathologists’ workflow. Bigpicture released tools for conversion of existing file formats to the DICOM standard, benefitting producers and users of AI in pathology. We also established a comprehensive metadata standard which strongly facilitates interoperability when adopted in the field.
Establishing datasets at partners’ sites for sharing within Bigpicture is complex, while required expertise is scarce. We produced workflows and tools to support high-quality data/metadata collection. Specifically, quality management tools and procedures were developed to ensure high-quality WSI. Software was developed to identify cases, extract data, convert images and prepare the cohort for easy upload to Bigpicture. These tools will enable access to large amounts of data that were so far locked in local institutions.
The EU landscape of data sharing regulations is variable and complex, resulting in an overly protective attitude, limiting availability of data. We published a report on GDPR interpretation and law for 8 EU countries, serving as a basis for creating local GDPR-compliant data sharing procedures. Availability of data for research and trust in the platform are thus increased, while adhering to all applicable laws and regulations.
The prospect of availability of large datasets created interest with commercial parties outside Bigpicture. With our beneficiaries, they will be instrumental in translating data and models into products/services creating tangible value in research and healthcare. We actively keep parties connected via webinars and conferences. Unforeseen use of Bigpicture data will be further explored by interviewing stakeholders to refine business models. We also initiated discussions with regulators on the use of digital pathology in a GLP environment, resulting in more efficient workflows and reduction of animal use in toxicity studies.
Beyond the direct project impact, Bigpicture serves as a pre-competitive environment to foster collaboration on shared themes such as quality management, regulatory challenges and future AI development thereby advancing the field of digital pathology.
Progress in this field is limited by a lack of standards, severely limiting interoperability and integration of AI-tools in the pathologists’ workflow. Bigpicture released tools for conversion of existing file formats to the DICOM standard, benefitting producers and users of AI in pathology. We also established a comprehensive metadata standard which strongly facilitates interoperability when adopted in the field.
Establishing datasets at partners’ sites for sharing within Bigpicture is complex, while required expertise is scarce. We produced workflows and tools to support high-quality data/metadata collection. Specifically, quality management tools and procedures were developed to ensure high-quality WSI. Software was developed to identify cases, extract data, convert images and prepare the cohort for easy upload to Bigpicture. These tools will enable access to large amounts of data that were so far locked in local institutions.
The EU landscape of data sharing regulations is variable and complex, resulting in an overly protective attitude, limiting availability of data. We published a report on GDPR interpretation and law for 8 EU countries, serving as a basis for creating local GDPR-compliant data sharing procedures. Availability of data for research and trust in the platform are thus increased, while adhering to all applicable laws and regulations.
The prospect of availability of large datasets created interest with commercial parties outside Bigpicture. With our beneficiaries, they will be instrumental in translating data and models into products/services creating tangible value in research and healthcare. We actively keep parties connected via webinars and conferences. Unforeseen use of Bigpicture data will be further explored by interviewing stakeholders to refine business models. We also initiated discussions with regulators on the use of digital pathology in a GLP environment, resulting in more efficient workflows and reduction of animal use in toxicity studies.
Beyond the direct project impact, Bigpicture serves as a pre-competitive environment to foster collaboration on shared themes such as quality management, regulatory challenges and future AI development thereby advancing the field of digital pathology.