Periodic Reporting for period 3 - INCISIVE (A multimodal AI-based toolbox and an interoperable health imaging repository for the empowerment of imaging analysis related to the diagnosis, prediction and follow-up of cancer)
Okres sprawozdawczy: 2023-04-01 do 2024-03-31
INCISIVE, a 42-month project, aimed at exploring the potential of novel AI tools for enhancing current imaging solutions for cancer cases thus supporting more effective decision-making for Healthcare Professionals. The project also aimed at delivering a reusable big data repository of cancer images and clinical data that can be used for training and validating AI tools for imaging methods.
In the end of the project, INCISIVE delivered an integrated platform. Depending on the role of the registered and authorized user, the functionalities that can be accessed through the platform include: 1) a federated cancer image data repository, enabling GDPR-compliant federated data search, access and management functionalities requiring prior registration; 2) AI training-related functionalities, including federated learning of AI models using distributed interoperable data; 3) explainable, AI-enabled decision-support services delivered to Health care professionals (HCPs); 4) a public section presenting all relevant information about the INCISIVE data sharing mechanism and supporting tools, and promoting data sharing among data holders.
The AI services were piloted on four types of cancer: lung, colorectal, breast and prostate cancer with extensive pilot activities carried out in 8 sites across 5 countries Greece, Italy, Spain, Cyprus and Serbia.
In the end of the project, INCISIVE delivered an integrated platform. Depending on the role of the registered and authorized user, the functionalities that can be accessed through the platform include: 1) a federated cancer image data repository, enabling GDPR-compliant federated data search, access and management functionalities requiring prior registration; 2) AI training-related functionalities, including federated learning of AI models using distributed interoperable data; 3) explainable, AI-enabled decision-support services delivered to Health care professionals (HCPs); 4) a public section presenting all relevant information about the INCISIVE data sharing mechanism and supporting tools, and promoting data sharing among data holders.
The AI services were piloted on four types of cancer: lung, colorectal, breast and prostate cancer with extensive pilot activities carried out in 8 sites across 5 countries Greece, Italy, Spain, Cyprus and Serbia.
The work carried out throughout the project lifetime concluded to the following main results:
1. Τhe final version of the AI toolbox comprising AI services offering decision-support for all 4 cancer types addressed by the project and combining several AI models per AI service. The main decision-support services offered relate to classification of abnormalities, patient prioritization, lesion segmentation and localization assistance, cancer diagnosis and staging and risk for metastasis prediction. The AI toolbox includes a total of 28 AI models and is integrated with a set of intuitive and interactive visualisation and reporting tools that support usability of the AI services, facilitating the delivery of the AI tools’ outcomes to healthcare professionals, such as image-to-report transformation, explainable AI features, the implemented approach on federated learning, and AR-enhanced visualizations.
2. The final version of the INCISIVE federated data repository, supporting the data sharing of more than 4 million interoperable cancer images and accompanying clinical data from 9 data providers in a GDPR-compliant way. The repository allows health data sharing among registered stakeholders in compliance with legal, ethical, privacy and security requirements, for AI-related training and research experimentation. The repository aggregates anonymised cancer image data and accompanying clinical data from more than 10.000 individuals. It relies on federated data storage abiding to high data privacy and security standards, allowing data sharing through a central node and 5 distributed federated nodes located at the data holders’ site, depending on data holders’ preference. The repository is extensible and scalable and can easily integrate additional federated data nodes.
3. The final version of the INCISIVE integrated platform, integrating all planned functionalities and making them available to end users, following appropriate authorisation depending on their role (data provider, data user, AI services user).
4. A GDPR-compliant data sharing mechanism, covering both technical and operational aspects of data sharing, and complying with legal and ethical norms before data is shared. This mechanism supports interested data providers to share their data within the INCISIVE hybrid/federated repository and allows data users to search and reuse INCISIVE data, respecting the respective data access rights imposed by data holders.
5. A federated learning mechanism, which enables the training of AI models by leveraging distributed data stored in different locations.
6. A data interoperability framework including a methodology for data integration, as well as a documented DICOM- and FHIR-based Common Data Model and tools supporting integration of heterogeneous cancer image and clinical data from multiple data providers.
Extensive dissemination and exploitation activities have been carried out to bring out the full potential of the aforementioned project results. All project results will be sustained, made accessible for further reuse and extension through the cancer image digital infrastructure (https://cancerimage.eu/) co-funded by the EC, that is currently being implemented by the EUCAIM project.
1. Τhe final version of the AI toolbox comprising AI services offering decision-support for all 4 cancer types addressed by the project and combining several AI models per AI service. The main decision-support services offered relate to classification of abnormalities, patient prioritization, lesion segmentation and localization assistance, cancer diagnosis and staging and risk for metastasis prediction. The AI toolbox includes a total of 28 AI models and is integrated with a set of intuitive and interactive visualisation and reporting tools that support usability of the AI services, facilitating the delivery of the AI tools’ outcomes to healthcare professionals, such as image-to-report transformation, explainable AI features, the implemented approach on federated learning, and AR-enhanced visualizations.
2. The final version of the INCISIVE federated data repository, supporting the data sharing of more than 4 million interoperable cancer images and accompanying clinical data from 9 data providers in a GDPR-compliant way. The repository allows health data sharing among registered stakeholders in compliance with legal, ethical, privacy and security requirements, for AI-related training and research experimentation. The repository aggregates anonymised cancer image data and accompanying clinical data from more than 10.000 individuals. It relies on federated data storage abiding to high data privacy and security standards, allowing data sharing through a central node and 5 distributed federated nodes located at the data holders’ site, depending on data holders’ preference. The repository is extensible and scalable and can easily integrate additional federated data nodes.
3. The final version of the INCISIVE integrated platform, integrating all planned functionalities and making them available to end users, following appropriate authorisation depending on their role (data provider, data user, AI services user).
4. A GDPR-compliant data sharing mechanism, covering both technical and operational aspects of data sharing, and complying with legal and ethical norms before data is shared. This mechanism supports interested data providers to share their data within the INCISIVE hybrid/federated repository and allows data users to search and reuse INCISIVE data, respecting the respective data access rights imposed by data holders.
5. A federated learning mechanism, which enables the training of AI models by leveraging distributed data stored in different locations.
6. A data interoperability framework including a methodology for data integration, as well as a documented DICOM- and FHIR-based Common Data Model and tools supporting integration of heterogeneous cancer image and clinical data from multiple data providers.
Extensive dissemination and exploitation activities have been carried out to bring out the full potential of the aforementioned project results. All project results will be sustained, made accessible for further reuse and extension through the cancer image digital infrastructure (https://cancerimage.eu/) co-funded by the EC, that is currently being implemented by the EUCAIM project.
The impact of the aforementioned results is high and has already started to be realized.
By offering decision-support assistance to medical professionals, the novel AI services of INCISIVE can have a positive impact on the clinical workflow, e.g. by supporting medical professionals in faster diagnosis.
Importantly, the millions of longitudinal cancer image and accompanying clinical data in the INCISIVE repository can be used as benchmark data for the training and validation of many more AI models to be implemented in the future by interested researchers.
The project’s impact is further amplified through the participation of INCISIVE to the AI for Health Imaging (AI4HI) cluster of projects and network (https://ai4hi.net/ ). The AI4HI cluster is an active network that brings together 7 projects working on AI and cancer image repositories and includes more than 130 organisations covering almost the whole of Europe. INCISIVE is actively contributing with the know-how derived from the project to discussions on how to address common challenges such as data interoperability, data de-identification, data storage, AI development and validation, etc.
Moreover, INCISIVE members are leading a book on Trustworthy AI, expected to be published in 2024, which gathers findings and lessons learned from INCISIVE and other AI4HI cluster project activities.
Last but not least, through multiple past and future activities, including interaction with the EC and multi-stakeholder initiatives, INCISIVE has and will continue to feed into policy and standardization efforts with impact on the European Health Data Space (EHDS).
By offering decision-support assistance to medical professionals, the novel AI services of INCISIVE can have a positive impact on the clinical workflow, e.g. by supporting medical professionals in faster diagnosis.
Importantly, the millions of longitudinal cancer image and accompanying clinical data in the INCISIVE repository can be used as benchmark data for the training and validation of many more AI models to be implemented in the future by interested researchers.
The project’s impact is further amplified through the participation of INCISIVE to the AI for Health Imaging (AI4HI) cluster of projects and network (https://ai4hi.net/ ). The AI4HI cluster is an active network that brings together 7 projects working on AI and cancer image repositories and includes more than 130 organisations covering almost the whole of Europe. INCISIVE is actively contributing with the know-how derived from the project to discussions on how to address common challenges such as data interoperability, data de-identification, data storage, AI development and validation, etc.
Moreover, INCISIVE members are leading a book on Trustworthy AI, expected to be published in 2024, which gathers findings and lessons learned from INCISIVE and other AI4HI cluster project activities.
Last but not least, through multiple past and future activities, including interaction with the EC and multi-stakeholder initiatives, INCISIVE has and will continue to feed into policy and standardization efforts with impact on the European Health Data Space (EHDS).