Mask-Free Neuron Concept Annotation for Interpreting Neural Networks in Medical Domain

TL;DR

This paper introduces MAMMI, a novel mask-free method using vision-language models to interpret neural networks in medical imaging, reducing reliance on pixel-wise annotations and enhancing transparency in clinical AI applications.

Contribution

MAMMI is the first approach to interpret medical neural networks without pixel-level masks, leveraging vision-language models for richer neuron representations.

Findings

MAMMI outperforms existing interpretation methods in medical image analysis.

Experiments on NIH chest X-rays validate MAMMI's effectiveness for clinical decision transparency.

The method reduces annotation costs by eliminating the need for pixel-wise masks.

Abstract

Recent advancements in deep neural networks have shown promise in aiding disease diagnosis and medical decision-making. However, ensuring transparent decision-making processes of AI models in compliance with regulations requires a comprehensive understanding of the model's internal workings. However, previous methods heavily rely on expensive pixel-wise annotated datasets for interpreting the model, presenting a significant drawback in medical domains. In this paper, we propose a novel medical neuron concept annotation method, named Mask-free Medical Model Interpretation (MAMMI), addresses these challenges. By using a vision-language model, our method relaxes the need for pixel-level masks for neuron concept annotation. MAMMI achieves superior performance compared to other interpretation methods, demonstrating its efficacy in providing rich representations for neurons in medical image analysis. Our experiments on a model trained on NIH chest X-rays validate the effectiveness of MAMMI, showcasing its potential for transparent clinical decision-making in the medical domain. The code is available at https://github.com/ailab-kyunghee/MAMMI.