Med-SegLens: Latent-Level Model Diffing for Interpretable Medical Image Segmentation

TL;DR

Med-SegLens introduces a latent-level model diffing framework that interprets segmentation models, identifies dataset shifts, and enables targeted interventions to improve performance without retraining.

Contribution

This work presents Med-SegLens, a novel approach for decomposing segmentation model activations into interpretable latents, facilitating diagnosis and correction of failures across datasets.

Findings

Identifies stable shared representations across models and datasets.

Latent interventions can recover 70% of failure cases.

Improves Dice score from 39.4% to 74.2% after interventions.

Abstract

Modern segmentation models achieve strong predictive performance but remain largely opaque, limiting our ability to diagnose failures, understand dataset shift, or intervene in a principled manner. We introduce Med-SegLens, a model-diffing framework that decomposes segmentation model activations into interpretable latent features using sparse autoencoders trained on SegFormer and U-Net. Through cross-architecture and cross-dataset latent alignment across healthy, adult, pediatric, and sub-Saharan African glioma cohorts, we identify a stable backbone of shared representations, while dataset shift is driven by differential reliance on population-specific latents. We show that these latents act as causal bottlenecks for segmentation failures, and that targeted latent-level interventions can correct errors and improve cross-dataset adaption without retraining, recovering performance in 70% of failure cases and improving Dice score from 39.4% to 74.2%. Our results demonstrate that latent-level model diffing provides a practical and mechanistic tool for diagnosing failures and mitigating dataset shift in segmentation models.