CHAIMELEON Project: Creation of a Pan-European Repository of Health Imaging Data for the Development of AI-Powered Cancer Management Tools

Luis Martí‐Bonmatí, Ana Miguel, A. Suárez-Garcı́a, Mario Aznar, Jean Paul Beregi, Laure Fournier, Emanuele Neri, Andrea Laghi, Manuela França, Francesco Sardanelli, Tobias Penzkofer, Philippe Lambin, Ignácio Blanquer, Marion I. Menzel, Karine Seymour, Sergio Figueiras Gómez, Katharina Krischak, Ricard Martínez Martínez, Yisroel Mirsky, Guang Yang, Ángel Alberich‐BayarriFrontiers in Oncology 2022 · DOI

TL;DR

The CHAIMELEON project is building a pan-European, FAIR (Findable, Accessible, Interoperable, Reusable) data repository to accelerate AI development for cancer management. By aggregating diverse imaging and clinical data, along with standardized processing tools, it aims to overcome current limitations in dataset availability and interoperability, paving the way for more robust and clinically relevant AI solutions.

What Problem Does This Paper Address?

Developing effective AI tools for cancer management is severely hampered by the lack of large, high-quality, and standardized datasets. Current challenges include:

  • Data Scarcity and Fragmentation: Imaging data is often siloed within individual institutions, making it difficult to gather sufficient diverse data for robust AI training.
  • Interoperability Issues: Different imaging equipment, software versions, and data formats across institutions lead to inconsistencies, making it hard to combine and compare data.
  • Data Harmonization: Quantitative imaging biomarkers (QIBs) can vary significantly between scanners and institutions, hindering reproducibility and reliable analysis.
  • Legal and Ethical Hurdles: Navigating data privacy regulations (like GDPR) and ensuring ethical data use for research is complex.
  • Lack of Standardized Tools: A unified environment for AI experimentation, including data processing, annotation, and validation, is missing.

CHAIMELEON aims to address these by creating a centralized, yet distributed, repository that standardizes data, processing, and access, thereby fostering AI innovation in oncology.

Methods

CHAIMELEON employs a multi-faceted approach:

  • Distributed Cloud-Based Repository: The core infrastructure is cloud-based, utilizing open standards and services (like EOSC, OpenStack, Docker, Kubernetes) for scalability and cost-efficiency. It features a hybrid architecture with local data warehouses at hospitals feeding into a central repository.
  • Data Processing Pipelines: Automated pipelines are being developed for:
  • De-identification and Pseudonymization: Ensuring patient privacy at multiple levels.
  • Curation and Annotation: Standardizing and enriching the data.
  • Image Harmonization: Developing novel methods, including GAN-based models, to ensure consistency across diverse imaging sources and enable reproducible QIB extraction.
  • Data Authenticity Protection: Securing the integrity of processed data.
  • FAIR Data Principles: The repository is designed to be Findable, Accessible, Interoperable, and Reusable, using standards like DICOM and MIABIS for metadata.
  • Controlled Access Policy: A tiered access system will be implemented, requiring user registration and adherence to data usage terms.
  • AI-Powered Analytical Engine: This engine will facilitate the interpretation, extraction, and exploitation of information from the repository.
  • Validation Framework: The project includes technical validation through internal testing and external "Open Challenges" for the AI community, followed by clinical validation via in-silico studies.

Datasets

The CHAIMELEON repository will be populated with retrospective, real-world patient data from clinical partners and external collaborators. The initial focus is on four high-prevalence cancer types:

  • Lung Cancer: CT/PET/CT imaging.
  • Breast Cancer: Mammography, Digital Breast Tomosynthesis, Ultrasound, and MRI.
  • Colorectal Cancer: Colon (CT) and Rectum (MRI) imaging.
  • Prostate Cancer: MRI imaging.

The project forecasts an initial dataset of approximately 11,500 cases of lung cancer, 6,000 of breast cancer, 6,000 of colorectal cancer, and 10,000 of prostate cancer, growing to nearly 40,000 cases (around 20 million images) by project end with external collaborators.

Key Insights and Ideas

  • The "Ecosystem" Approach: Recognizing that no single institution can solve the data problem, CHAIMELEON builds a collaborative ecosystem of 18 European partners (hospitals, universities, R&D centers) to pool resources and expertise.
  • Beyond Data Storage: A Processing and Validation Platform: The repository isn't just a data dump; it's designed as an active platform with integrated tools for data processing, harmonization, and AI model validation, accelerating the research lifecycle.
  • Focus on Harmonization as a Core Challenge: The paper emphasizes that simply collecting data isn't enough; making it quantitatively comparable through advanced harmonization techniques is crucial for reliable AI.
  • Bridging the Gap to Clinical Utility: The project explicitly plans for technical and clinical validation, ensuring that the developed AI tools are not just technically sound but also clinically relevant and useful for addressing specific clinical end points (CEPs).
  • Promoting FAIR Data and Open Science: By adhering to FAIR principles and planning for open challenges, CHAIMELEON aims to contribute to a more open and collaborative research environment in medical imaging AI.

Key Results

As the project is ongoing, specific quantitative results are not yet finalized. However, the paper outlines expected outcomes:

  • A functional, pan-European, FAIR-compliant data repository for cancer imaging.
  • Development and validation of AI-powered pipelines for de-identification, curation, annotation, harmonization, and integrity securing.
  • A framework for legitimate access to anonymized imaging and clinical data for secondary research use across the EU.
  • Technical validation of the repository's usability and performance for AI experimentation.
  • Clinical validation of AI tools for assisting clinicians in addressing specific clinical end points (CEPs) for lung, breast, colorectal, and prostate cancer.
  • Contribution to the standardization of radiological procedures and imaging biomarker extraction.
  • Establishment of ethical standards for the use of health imaging data in AI development.

Limitations and Future Work

  • Data Inhomogeneity: While harmonization methods are being developed, the inherent variability in data from multiple sources remains a significant challenge.
  • Legal and Ethical Frameworks: The definition and implementation of data usage policies and contracting non-identification commitments are ongoing processes.
  • Scalability and Sustainability: While designed for scalability, long-term sustainability beyond the project's funding period will require ongoing assessment and potential synergies with other initiatives.
  • Future Work: The project aims to expand the repository to include data for other cancer types and imaging modalities, further enhancing its versatility and scalability. The developed methodologies and protocols are intended to be reusable for other research fields beyond oncology. The project also anticipates contributing to the evolution of data standards and promoting cooperation and data sharing among users.

Summary updated

Abstract

The CHAIMELEON project aims to set up a pan-European repository of health imaging data, tools and methodologies, with the ambition to set a standard and provide resources for future AI experimentation for cancer management. The project is a 4 year long, EU-funded project tackling some of the most ambitious research in the fields of biomedical imaging, artificial intelligence and cancer treatment, addressing the four types of cancer that currently have the highest prevalence worldwide: lung, breast, prostate and colorectal. To allow this, clinical partners and external collaborators will populate the repository with multimodality (MR, CT, PET/CT) imaging and related clinical data. Subsequently, AI developers will enable a multimodal analytical data engine facilitating the interpretation, extraction and exploitation of the information stored at the repository. The development and implementation of AI-powered pipelines will enable advancement towards automating data deidentification, curation, annotation, integrity securing and image harmonization. By the end of the project, the usability and performance of the repository as a tool fostering AI experimentation will be technically validated, including a validation subphase by world-class European AI developers, participating in Open Challenges to the AI Community. Upon successful validation of the repository, a set of selected AI tools will undergo early <i>in-silico</i> validation in observational clinical studies coordinated by leading experts in the partner hospitals. Tool 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. The project brings together a consortium of 18 European partners including hospitals, universities, R&D centers and private research companies, constituting an ecosystem of infrastructures, biobanks, AI<i>/in-silico</i> experimentation and cloud computing technologies in oncology.
BibTeX
@inproceedings{martibonmati2022chaimeleon,
  author = {Luis Martí‐Bonmatí and Ana Miguel and A. Suárez-Garcı́a and Mario Aznar and Jean Paul Beregi and Laure Fournier and Emanuele Neri and Andrea Laghi and Manuela França and Francesco Sardanelli and Tobias Penzkofer and Philippe Lambin and Ignácio Blanquer and Marion I. Menzel and Karine Seymour and Sergio Figueiras Gómez and Katharina Krischak and Ricard Martínez Martínez and Yisroel Mirsky and Guang Yang and Ángel Alberich‐Bayarri},
  title = {{CHAIMELEON Project: Creation of a Pan-European Repository of Health Imaging Data for the Development of AI-Powered Cancer Management Tools}},
  booktitle = {Frontiers in Oncology},
  year = {2022},
  doi = {10.3389/fonc.2022.742701},
}
Figure