Torus presents article on Dermatology and AI at PFIA 2024
PhD Marianne Defresne and PhD Paul Fricker from Torus presented their article on the explainability of their skin lesion classification AI at the PFIA 2024 Health & AI Day. PFIA, which stands for Plate-Forme Intelligence Artificielle, is an annual event for AI researchers in France.
During an engaging session, Marianne and Paul interacted with other researchers and doctors who were interested in their work on Explainable AI and Segmentation.
Their article discusses the success of Deep Learning in detecting skin cancer from lesion images, highlighting the challenge of the lack of transparency in its decision-making process. The model they developed is no exception. To address this, they propose two approaches based on dermatology rules for lesion diagnosis:
- Visualization Tool for Contextual Understanding: They introduce a visualization tool designed to provide practitioners with additional context for their model’s decisions. This tool highlights important regions of the image, helping practitioners verify and understand the model’s focus areas.
- Medical Concept-Based Model Variant: They present a variant of their baseline model that incorporates medical concepts derived from established dermatology rules. By integrating domain-specific knowledge into the model’s architecture or training process, this approach enhances the interpretability and alignment with medical practices.
Learn more about their article here
