Periodic Reporting for period 2 - EuCanImage (A European Cancer Image Platform Linked to Biological and Health Data for Next-Generation Artificial Intelligence and Precision Medicine in Oncology)
Okres sprawozdawczy: 2022-04-01 do 2023-09-30
Europe has been at the forefront of AI developments in cancer imaging as illustrated by the introduction of the radiomics concept in 2012 by members of the EuCanImage consortium. However, clinical translation of existing AI-based cancer imaging solutions is still lacking, as all-too-often, they have been built and validated in small site-specific cancer imaging datasets.
To promote trust, clinical value and translation of emerging AI technologies for cancer imaging, not only larger and more diverse data sets are needed, but also new standards and improved practices to exploit such large datasets for addressing clinical needs that are currently unmet in oncology. Nevertheless, as large cancer imaging repositories are still missing in Europe, access to multi-centre cancer imaging dataset for the AI communities and industries remains a challenge. Thus, EuCanImage responds to an imperative need for such high-quality and large repositories in Europe to leverage the available expertise in AI, radiomics and cancer imaging for precision medicine in oncology.
Cancer imaging is a central piece to realise the promise of precision medicine in oncology. Nonetheless, non-image data are also relevant, particularly -omics and health data, for enabling to build integrated multi-scale AI models that consider patient-specific biomolecular, phenotypic, environmental and clinical information. It is expected that integrated AI models will lead to improved diagnosis and treatment selection for many cancer types. Hence, it is important to develop cancer imaging repositories that are cross-linked to corresponding -omics and health datasets.
The main objective of EuCanImage is to build and demonstrate a General Data Protection Regulation (GDPR)-compliant and scalable AI platform for leveraging large-scale, high-quality and interoperable cancer imaging datasets adequately linked to biological and health oncology data. This user-friendly platform will provide functionalities that enable easy access to information on available data and facilitate future data depositions. This platform will also integrate advanced capabilities and new standards to develop and validate AI-powered clinical decision support systems for precision oncology with increased clinical trust and adoption, while developing the legal framework that will enable responsible data sharing and enhanced Open Science.
To promote trust, clinical value and translation of emerging AI technologies for cancer imaging, not only larger and more diverse data sets are needed, but also new standards and improved practices to exploit such large datasets for addressing clinical needs that are currently unmet in oncology. Nevertheless, as large cancer imaging repositories are still missing in Europe, access to multi-centre cancer imaging dataset for the AI communities and industries remains a challenge. Thus, EuCanImage responds to an imperative need for such high-quality and large repositories in Europe to leverage the available expertise in AI, radiomics and cancer imaging for precision medicine in oncology.
Cancer imaging is a central piece to realise the promise of precision medicine in oncology. Nonetheless, non-image data are also relevant, particularly -omics and health data, for enabling to build integrated multi-scale AI models that consider patient-specific biomolecular, phenotypic, environmental and clinical information. It is expected that integrated AI models will lead to improved diagnosis and treatment selection for many cancer types. Hence, it is important to develop cancer imaging repositories that are cross-linked to corresponding -omics and health datasets.
The main objective of EuCanImage is to build and demonstrate a General Data Protection Regulation (GDPR)-compliant and scalable AI platform for leveraging large-scale, high-quality and interoperable cancer imaging datasets adequately linked to biological and health oncology data. This user-friendly platform will provide functionalities that enable easy access to information on available data and facilitate future data depositions. This platform will also integrate advanced capabilities and new standards to develop and validate AI-powered clinical decision support systems for precision oncology with increased clinical trust and adoption, while developing the legal framework that will enable responsible data sharing and enhanced Open Science.
The EuCanImage project made significant advancements, starting by addressing legal and ethical aspects and establishing guidelines for sharing oncologic imaging data for AI, as part of Work Packages 1 and 9. The project completed the requirements analysis for all its use cases, encompassing a range of elements like case descriptions, anonymisation methods, and data linkage protocols. Around 9,000 images from our clinical partners are currently being segmented on the CMRAD platform (Work Package 2), and we've achieved improved integration between CMRAD and the XNAT repository, streamlining data imports. In Work Package 3, we've developed a FAIR-compliant data point within XNAT for managing non-raw DICOM data, which is poised for further expansion. The coding frameworks and data models for non-imaging clinical data have been finalised, and electronic case report forms (eCRFs) are now operational for uploading non-imaging data. We've also established linkages between XNAT, EGA (based in Europe), and the project catalogue, including a connection to TCIA (based in the US), to support international studies. Work Package 4 focused on optimising data curation and quality control, tailoring these processes to specific use cases. Annotation strategies for colorectal and liver cases are in place, and tools for breast case annotations, utilising public datasets, have been integrated into the AI-VRE, which now includes the completed Machine Learning Toolbox (T5.3). Work Package 5 saw the introduction of an updated, publicly available version of the Federated Learning platform and the implementation of the latest FUTURE-AI guidelines, reflecting insights from over 100 experts in multiple fields. The project's in-silico trial platform was implemented in Work Package 6 to enhance AI evaluations, while discussions are ongoing to incorporate metrics into the OpenEBench tool for AI solution benchmarking. Communication and dissemination efforts under Work Package 7 have resulted in enhanced website performance, increased social media engagement, and successful video content with an emphasis on data annotations. The overall dissemination strategy has been thoroughly reviewed and updated in Deliverable D7.3. The project coordinator continues to provide scientific, administrative and financial support to all partners, while encouraging EuCanImage partners to engage with other initiatives, such as the AI4HI network, through joint events and papers.
EuCanImage will constitute the first repository of cancer imaging data, integrating information on imaging sequences, available annotations and imaging biomarkers, and directly cross-linked to biological and health repositories. A legal governance framework will be established to promote data sharing for the benefit of science and society, while ensuring compliance with regulation on data privacy and protecting the rights of the citizens. Various tools will be delivered to enable data anonymisation, data curation, clinical annotations and secure storage in the most secure European infrastructures, namely Euro-Bioimaging and European Genome-phenome Archive. An AI environment will be developed with integrated capabilities in machine learning, federated learning, radiomics and explainable AI, to build the next generation of multi-modal AI solutions in cancer. The EuCanImage platform as a whole is expected to impact oncology research by creating an EU-wide repository of health images dedicated to the most common forms of cancer, leading to an improvement in diagnosis, treatment and follow-up and contributing to a more precise and personalised management of cancer via AI-based solutions. Datasets of EuCanImage have been carefully selected to constitute a unique resource currently unavailable in existing repositories, thereby allowing AI experts and clinical stakeholders to investigate unmet clinical needs on different types of cancers, especially for breast, liver and colorectal cancer. AI-based diagnostic and treatment planning tools will be validated on conditions that have huge personal, social and economic costs, resulting in improved outcomes for citizens and the society as a whole. EuCanImage will also develop standards and best practices for designing, developing and validating future AI solutions which will be technically robust, clinical safe and free of bias, hence they will be trusted, deployed and exploited in real-world cancer care. Additionally, these innovations will benefit the field of imaging AI in oncology, but also medical imaging at large, bringing the quality of the implemented image-data catalogues to the next level and highest standard for future medical imaging in Europe.