Mammo-SAE: Interpreting Breast Cancer Concept Learning with Sparse Autoencoders

TL;DR

This paper introduces a Sparse Autoencoder-based interpretability method for breast imaging models, revealing how latent features relate to clinical concepts and decision factors in mammogram analysis.

Contribution

It presents a novel application of Sparse Autoencoders to interpret foundation models in breast imaging, linking latent features to clinical concepts and decision influences.

Findings

Latent neurons often align with ground truth regions.

Identifies confounding factors affecting model decisions.

Analyzes neuron reliance during downstream fine-tuning.

Abstract

Interpretability is critical in high-stakes domains such as medical imaging, where understanding model decisions is essential for clinical adoption. In this work, we introduce Sparse Autoencoder (SAE)-based interpretability to breast imaging by analyzing {Mammo-CLIP}, a vision--language foundation model pretrained on large-scale mammogram image--report pairs. We train a patch-level \texttt{Mammo-SAE} on Mammo-CLIP to identify and probe latent features associated with clinically relevant breast concepts such as \textit{mass} and \textit{suspicious calcification}. Our findings reveal that top activated class level latent neurons in the SAE latent space often tend to align with ground truth regions, and also uncover several confounding factors influencing the model's decision-making process. Additionally, we analyze which latent neurons the model relies on during downstream finetuning for improving the breast concept prediction. This study highlights the promise of interpretable SAE latent representations in providing deeper insight into the internal workings of foundation models at every layer for breast imaging. The code will be released at https://krishnakanthnakka.github.io/MammoSAE/