GRASPing Anatomy to Improve Pathology Segmentation

TL;DR

GRASP is a modular framework that enhances pathology segmentation by integrating anatomical context through pseudolabels and feature alignment, improving accuracy without retraining anatomical models.

Contribution

Introduces GRASP, a plug-and-play method that leverages existing anatomy models to improve pathology segmentation without retraining anatomical components.

Findings

Consistently achieves top rankings across datasets and architectures.

Effectively incorporates anatomical context through dual injection strategy.

Improves segmentation accuracy by leveraging anatomical information.

Abstract

Radiologists rely on anatomical understanding to accurately delineate pathologies, yet most current deep learning approaches use pure pattern recognition and ignore the anatomical context in which pathologies develop. To narrow this gap, we introduce GRASP (Guided Representation Alignment for the Segmentation of Pathologies), a modular plug-and-play framework that enhances pathology segmentation models by leveraging existing anatomy segmentation models through pseudolabel integration and feature alignment. Unlike previous approaches that obtain anatomical knowledge via auxiliary training, GRASP integrates into standard pathology optimization regimes without retraining anatomical components. We evaluate GRASP on two PET/CT datasets, conduct systematic ablation studies, and investigate the framework's inner workings. We find that GRASP consistently achieves top rankings across multiple evaluation metrics and diverse architectures. The framework's dual anatomy injection strategy, combining anatomical pseudo-labels as input channels with transformer-guided anatomical feature fusion, effectively incorporates anatomical context.