Étude de la Dynamique Spatio-Temporelle Cellulaire par Spectroscopie d'Impédance sur des Réseau Multi-Électrodes Améliorée par Apprentissage Automatique

Monitoring cellular spatio-temporal dynamics is essential for understanding complex biological processes including organ development and cancer progression. Traditional methods such as live-cell fluorescence microscopy are limited by dye toxicity and photo-induced cell damage. To address these challenges, an innovative approach combining micro-electrode arrays (MEAs) with electrical impedance spectroscopy (EIS) and Machine Learning (ML) is proposed as a non-invasive, label-free and cost-effective alternative for the monitoring of cellular mixtures.The MEA design incorporates 25 electrode pairs in a central sensing region, surrounded by a cell culture well. A detailed microfabrication protocol is developed, along with a custom assembly and hardware setup for stable device connection. Mono-culture and co-culture experiments are conducted using normal (MCF10A) and cancerous (MCF7) breast epithelial cell lines. EIS measurements are combined with automatic cell segmentation of microscopy images to correlate electrical signals with cell growth parameters.A Long Short-Term Memory (LSTM) deep learning model is developed to predict cell density, covered area, mean cell diameter, and to classify cell types from the IS signals. The model demonstrates good performance in predicting image segmentation parameters for normal cells (R² > 0.7), with somewhat poorer performance for cancerous cells (R² > 0.4). EIS classification between normal and cancerous cells yields high accuracy in mono-culture (AUC 0.96), with lower scores in mixed media conditions (AUC 0.81) and co-culture (AUC 0.72).While challenges remain in addressing inter-experimental variability, this study presents the first quantitative dataset in the field, with over 30,000 EIS signals and corresponding image labels, that has been used to compare IS to multiple cellular parameters in a spatio-temporal manner. Additionally, the study provides a foundation in the methodology and tools to implement the technology in future research studies, holding clear potential in tissue engineering, microphysiological systems and bioreactor applications.