Farida ZEHRAOUI will defend her Habilitation à Diriger les Recherches on Friday February 23, 2024: “Interpretability of Deep Learning Models in Personalized Medicine”.

Within the realm of healthcare, machine learning (ML) techniques have demonstrated remarkable potential in enhancing diagnostics, treatment strategies, and overall patient care.
Deep Learning has particularly shown impressive results in diverse healthcare applications, notably in medical image analysis and recognition.

However, deep learning models are considered “black-box” systems. In healthcare, where interpretability is crucial, understanding how a model arrives at a diagnosis or treatment recommendation is imperative for ensuring trust and widespread adoption.

To address this challenge, we proposed several deep learning and hybrid architectures that are interpretable by design. Our approaches include the following strategies:

– Integrating Knowledge Graphs into the deep learning architectures, providing concept-based explanations.

– Combining Supervised deep learning with an unsupervised explainable model to provide prototype-based explanations.

– Combining Symbolic AI approaches with deep learning to ensure rule-based explanations.

These proposed advancements in explainable deep learning models aim to instill trust in model predictions, ensuring their beneficial use in the landscape of personalized medicine.