Artificial intelligence as a new tool for cancer management coordination
CHAIMELEON aims to set up an EU-wide structured repository for health imaging data to be openly reused in Artificial Intelligence (AI) experimentation for cancer management. It will be built as a distributed infrastructure and populated with multimodality imaging and related clinical data for historic and newly diagnosed lung, prostate and colorectal cancer patients.
The CHAIMELEON repository will be designed to meet the need for access to large datasets of high quality in interoperable repositories enabling secure sharing across Europe with the specific purpose of constituting an open, accessible, intuitive resource for the community of developers of AI-based solutions for cancer management.
Facts and Figures
| Name | Accelerating the lab to market transition of AI tools for cancer management |
| Acronym | CHAIMLEON |
| Start date | September 1, 2020 |
| End date | August 31, 2024 |
| Project Coordinator | Prof. Luis Marti-Bonmati, HULAFE |
| Consortium | 18 partners from 10 coutnries |
| Total funding | € 8 784 038,75 |
Context
The use of Artificial Intelligence (AI) on health data is generating promising tools for assisting clinicians in application fields such as cancer management. Increasing favourable outcomes suggests that health imaging-based AI approaches can become useful clinical tools in areas such as non-invasive tumour characterization, prediction of certain tumour features, staging of tumour spread, stratification of patients, selection of most appropriate therapies and clinical prognosis. Thus, the development and validation of health imaging AI tools is a field of great interest nowadays, not only from a research perspective but also from a legal (e.g. privacy protection and ethics), regulatory (e.g. CE marking, accountability for supported clinical decisions) and operational (e.g. standards, ontology, large repositories) perspective in the common interest of accelerating the path to market of these AI tools once they have proven to be useful, safe, legal and accepted by the clinical community.
The advancement of health imaging AI research is intrinsically linked to the availability of quality controlled large collected datasets. In the EU healthcare systems, data is collected in a comprehensive manner for each patient, all through the disease management cycle, which offers an excellent potential for the use of Big Data in cancer management. However, the access to large volume of curated datasets with controlled confounding factors remains as a major challenge. Ensuring that data are consistent and representative of the context of application (e.g. with clear established diagnosis criteria for patient inclusion) implies important trade‐offs between data quantity and data quality. The generation of quality large datasets is a resource-intensive endeavour, facing technical and operational difficulties such as data harmonisation, data curation and standardisation of annotation, as well as legal and ethical restrictions. These repositories are crucial for gathering real world evidence drawing causal inferences throughout non-interventional observational studies.
The CHAIMELEON Repository
The CHAIMELEON repository is expected to be used throughout the European Union as a common infrastructure that complies with all ethical and safety regulations of the countries involved. Its development will build on partner’s experience (e.g. PRIMAGE repository for paediatric cancer and the Euro-BioImaging node for Valencia population, by HULAFE; the Radiomics Imaging Archive by UM; the national repository DRIM AI France, the Oncology imaging biobank by UNIPI). Clinical partners and external collaborators will populate the Repository with multimodality (MR, CT, PET/CT) imaging and related clinical data for historic and newly diagnosed lung, prostate and colorectal cancer patients. The project aims to incorporate about 40,000 cases in the repository, which will be contributed by eight clinical partners. In addition, in order to demonstrate the universality of the models developed, external collaborators will be sought to contribute more medical images and clinical data to the common repository.
A multimodal analytical data engine will facilitate interpretation, extraction, and exploitation of the stored information. An ambitious development and implementation of AI-powered pipelines will enable advancement towards automating data deidentification, curation, annotation, integrity securing and images harmonisation, the latest being of the highest importance for enabling reproducibility of radiomics when using large multiscanner/multicentre image datasets.
In a comprehensive validation phase, the partners will ensure the usability and performance of the repository as a tool fostering AI experimentation. This will include a validation subphase by other world-class European AI developers articulated via the organisation of Open Challenges to the AI Community. A set of selected AI tools will undergo early on-silico validation in observational (non-interventional) clinical studies coordinated by leading experts at CERF (lung cancer), PSD (breast), ULS (colorectal) and HULAFE (prostate). Their performance will be assessed, including external independent validation, on hallmark clinical decisions in response to some of the currently most important clinical end points in cancer.
