Multimodal Medical Data Fusion Using Deep Learning for Diagnostic Support of Breast Cancer Subtypes

The DiMuCaSe-DL project aims to improve the diagnosis of breast cancer subtypes, particularly complex ones such as triple-negative breast cancer (TNBC), by combining multiple sources of medical data using advanced deep learning techniques. The project leverages the complementarity of several modalities: FTIR spectral imaging, mammography, histopathological analyses (H&E staining, immunohistochemistry), and clinical data. A retrospective cohort will be established, with at least 30 patients per subtype. Supervised models (XGBoost, CNN) will be developed for each modality and then integrated at various levels (feature-level, decision-level, and multimodal fusion). The goal is to develop an efficient diagnostic tool based on artificial intelligence. The project relies on close collaboration between the BioSpecT and IRMAIC research units, the Godinot Institute, and the Institute of Artificial Intelligence in Health (IIAS), bringing together expertise in biophysics, immuno-oncology, radiomics, and AI. Expected outcomes include scientific (publications, conferences), clinical (improved patient management), and long-term technological impacts (development of software or hardware tools). This is an ambitious, multidisciplinary project with high innovation potential.

This subject is a collaboration between the BioSpecT and IRMAIC units of the University of Reims Champagne Ardenne, France.

The BioSpecT (Translational BioSpectroscopy) unit is a multidisciplinary group composed of clinicians, biologists, physicists, and bioinformaticians. Its mission is to develop vibrational spectroscopy approaches (Raman scattering and infrared absorption) for the characterization of biological samples, such as cells, tissues, and biofluids. These research efforts aim to identify new types of biomarkers, validated for clinical transfer (from bench to bedside approach), which could help clinicians predict the progression of a disease or its response to treatment as early as possible. These approaches rely on a combination of Raman or infrared techniques with multivariate chemometric data analysis, enabling the identification of reliable indicators of pathological biological processes in a completely objective, reproducible, and automated manner, without the need for labeling, reagents, or specific sample preparation. The identification of spectroscopic markers also involves characterizing biological models of the tumor microenvironment.

IRMAIC (Immunoregulation in Autoimmune, Inflammatory Diseases and Cancer) is a research unit of the University of Reims Champagne-Ardenne. The unit’s scientific project focuses on a better understanding of the cellular and molecular mechanisms involved in the development of immune-mediated inflammatory diseases. We explore the immuno-inflammatory mechanisms involved in the pathologies we study, with the aim of proposing new strategies for biosurveillance, diagnosis, and treatment. The team project applies to several pathologies: (i) Autoimmune and inflammatory diseases: Bullous pemphigoid, systemic scleroderma, primary humoral immunodeficiency, chronic obstructive pulmonary disease (COPD), and respiratory infections caused by Respiratory Syncytial Virus (RSV) and SARS-CoV-2, (ii) Cancers: Triple-negative breast cancer, chronic lymphocytic leukemia, and lung cancers associated with COPD.

The proposed doctoral subject requires to collaborate with experts in different scientific domains (biophysics, chemometrics, medicine, biology). The candidate will have a university education in artificial intelligence, signal and image processing, applied mathematics, bioinformatics or chemometrics. The candidate must justify of a strong interest in medical applications and of strong skills in Python.