Back to All Events

Poly-CAM: High resolution class activation map for convolutional neural networks


  • Explainable AI (XAI) is increasingly crucial as deep learning advances, especially in medical applications.

  • Our objective is to enhance feature localization accuracy in AI predictions for medical diagnostics.

  • We propose integrating information from different neural network layers to create high-resolution Class Activation Maps.

  • Our approach achieves competitive faithfulness metrics and significantly improves the localization of specific features.

• Project type: research

• Team status: complete

• Initiator: Alexandre Englebert, MD

• Co-investigators: Olivier Cornu, MD & PhD, Christophe De Vleeschouwer, PhD

• Fundings: FNRS/FRIA grant to Alexandre Englebert for his PhD

• Project status: done

Source: https://arxiv.org/abs/2204.13359
Previous
April 1

Rise of large language models in neurosurgery
Next
July 20

Representation Bias in Global Healthcare: AI-depicted vs. real-world medical leadership