Data Science & AI

The Data Science & AI research unit focuses on optimizing data collection processes in clinical care settings, constructing expansive research databases like the IJB Breast Cancer research database, and implementing multicentric data collection frameworks. These efforts enable the aggregation of big data and facilitate federated learning techniques, preserving patient privacy. Additionally, the unit develops AI-driven applications, including auto-segmentation tools and quantitative image analysis pipelines, and focuses on quality assurance initiatives for reproducible radiomics image analysis. As a third pillar, we are dedicated to the development of prognostic tools that aid oncology experts in making informed clinical decisions. To achieve this, we focus on multi-modal data, including anatomical and molecular imaging, digital pathology, clinical data and genomics, bringing together the expertise from various departments and research laboratories within our hospital in large multidisciplinary research projects. The global ambition of the Data Science & AI research unit is to improve patient care and outcomes, through the implementation of a learning healthcare system and the application of artificial intelligence and big data analysis.

A selection of our projects:

Imaging Score in chemorefractory metastatic colorectal cancer

Managing metastatic colorectal cancer (mCRC) that doesn’t respond to chemotherapy represents a delicate balance between treatment effectiveness and toxicity. To support treatment decision making, current prognostic tools like the Colon Life nomogram often concentrate on general patient health or one diagnostic method, lacking comprehensive assessment. This study proposes the Imaging Score, utilizing clinical and imaging data from PET/CT scans, to predict the probability of survival within twelve weeks of starting last-line treatment for mCRC.

AI for radiotherapy target optimisation in early-stage breast cancer

Breast cancer often spreads through the lymphatic system, leading to metastases in regional lymph nodes, which are commonly included in post-operative elective radiotherapy (eRT) to target potential microscopic tumors. However, the extent of irradiated tissue correlates with radiation-induced side effects, necessitating a careful balance between treatment efficacy and minimizing harm. This project aims to personalize the Clinical Target Volume (CTV) for eRT in early breast cancer (eBC), optimizing the selection and geometric definitions of lymph node levels (LNLs) to potentially de-escalate radiation treatment safely for certain patients.

ARTEMIS: Artificial Intelligence for tailoring in early mammary cancer – individual systemic therapy

The ARTEMIS research project aims to enable safe de-escalation of systemic neo-/adjuvant breast cancer treatment across the three major subtypes by integrating artificial intelligence (AI) and multimodal data. Primary objectives include establishing the ‘IJB Breast Cancer Research Database,’ a comprehensive repository of real-world patient data at IJB, facilitating breakthroughs in breast cancer research. Additionally, the project seeks to develop a clinical decision support system using AI methods on digital pathology and imaging data to provide personalized treatment recommendations, while also exploring novel biomarkers for early breast cancer through AI-driven discovery frameworks.

Optimisation of Lu177-PSMA treatment in metastatic colorectal cancer

This project aims to address the limited treatment options for metastatic castration-resistant prostate cancer (mCRPC) by establishing a national registry to collect standardized clinical and imaging data from patients undergoing Lu-177 PSMA therapy. Rigorous quality assurance and standardization will be implemented to ensure consistent image analysis, while two AI models will be developed to automate lesion segmentation and predict treatment response. These models will integrate clinical, laboratory, and imaging features to generate baseline prognostic scores and early-assessment predictive scores, ensuring enhanced clinical decision-making while preserving patient privacy through federated learning.

Selected Publications:

Real-time motion management in MRI-guided radiotherapy: Current status and AI-enabled prospects.
E Lombardo, J Dhont, D Page, et al. Radiother Oncol, 109970. (2023) DOI:10.1016/j.radonc.2023.109970

Pre-trial quality assurance of diffusion-weighted MRI for radiomic analysis and the role of harmonisation
Z Paquier, SL Chao, G Bregni, et al. Phys Med, 130;138-146. (2022) DOI:10.1016/j.ejmp.2022.10.009

Radiomics software comparison using digital phantom and patient data: IBSI-compliance does not guarantee concordance of feature values
Z Paquier, SL Chao, A Acquisto, et al. Biomed Phys Eng Express, 8(6). (2022) DOI: 10.1088/2057-1976/ac8e6f

Machine learning model using F-18-FDG PET/CT-based biomarkers to support clinical decision making in advanced chemorefractory metastatic colorectal cancer.
J Dhont, H Levillain, T Guiot, et al. Eur J Nucl Med Mol Imaging, 49(1);S242-243. (2022) DOI: 10.1007/s00259-022-05924-4