TP-Seg: Task-Prototype Framework for Unified Medical Lesion Segmentation

2026-04-01

AI Summary

This work addresses the limitations of existing unified medical lesion segmentation approaches, wherein shared encoders often induce feature entanglement and gradient interference, thereby degrading lesion discriminability. To overcome this, the authors propose TP-Seg, a novel framework that balances shared and task-specific representations through task-conditional adapters and introduces learnable task prototypes as semantic anchors. These prototypes, integrated via a cross-attention mechanism, enable fine-grained modeling of foreground–background semantic relationships. TP-Seg further employs a dual-path expert architecture and a prototype-guided decoder to effectively disentangle multimodal, multitask feature representations. Extensive experiments across eight medical imaging modalities demonstrate that TP-Seg significantly outperforms specialized, general-purpose, and current unified methods, exhibiting superior generalization capability and clinical applicability.

Abstract

Building a unified model with a single set of parameters to efficiently handle diverse types of medical lesion segmentation has become a crucial objective for AI-assisted diagnosis. Existing unified segmentation approaches typically rely on shared encoders across heterogeneous tasks and modalities, which often leads to feature entanglement, gradient interference, and suboptimal lesion discrimination. In this work, we propose TP-Seg, a task-prototype framework for unified medical lesion segmentation. On one hand, the task-conditioned adapter effectively balances shared and task-specific representations through a dual-path expert structure, enabling adaptive feature extraction across diverse medical imaging modalities and lesion types. On the other hand, the prototype-guided task decoder introduces learnable task prototypes as semantic anchors and employs a cross-attention mechanism to achieve fine-grained modeling of task-specific foreground and background semantics. Without bells and whistles, TP-Seg consistently outperforms specialized, general and unified segmentation methods across 8 different medical lesion segmentation tasks covering multiple imaging modalities, demonstrating strong generalization, scalability and clinical applicability.

Problem

• unified medical lesion segmentation
• feature entanglement
• gradient interference
• task heterogeneity
• medical imaging modalities

Innovation

• task-prototype
• unified segmentation
• task-conditioned adapter
• prototype-guided decoder
• cross-attention