SegMate: Asymmetric Attention-Based Lightweight Architecture for Efficient Multi-Organ Segmentation

2026-02-27

AI Summary

This work proposes SegMate, an efficient 2.5D lightweight framework for multi-organ segmentation in medical images, addressing the high computational cost and deployment challenges of existing models in resource-constrained clinical settings. SegMate integrates an asymmetric attention mechanism, multi-scale feature fusion, slice positional encoding, and multi-task learning, built upon lightweight backbones such as EfficientNetV2-M and MambaOut-Tiny. Evaluated on TotalSegmentator, the model achieves a Dice score of 93.51% with only 295 MB peak GPU memory usage—reducing computational demands by 2.5× and memory consumption by 2.1× compared to baseline models. Furthermore, it demonstrates strong zero-shot cross-dataset generalization, attaining a Dice score of up to 89.35%, thereby effectively balancing efficiency, accuracy, and robustness.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Large Multimodal Models (LMMs)
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

State-of-the-art models for medical image segmentation achieve excellent accuracy but require substantial computational resources, limiting deployment in resource-constrained clinical settings. We present SegMate, an efficient 2.5D framework that achieves state-of-the-art accuracy, while considerably reducing computational requirements. Our efficient design is the result of meticulously integrating asymmetric architectures, attention mechanisms, multi-scale feature fusion, slice-based positional conditioning, and multi-task optimization. We demonstrate the efficiency-accuracy trade-off of our framework across three modern backbones (EfficientNetV2-M, MambaOut-Tiny, FastViT-T12). We perform experiments on three datasets: TotalSegmentator, SegTHOR and AMOS22. Compared with the vanilla models, SegMate reduces computation (GFLOPs) by up to 2.5x and memory footprint (VRAM) by up to 2.1x, while generally registering performance gains of around 1%. On TotalSegmentator, we achieve a Dice score of 93.51% with only 295MB peak GPU memory. Zero-shot cross-dataset evaluations on SegTHOR and AMOS22 demonstrate strong generalization, with Dice scores of up to 86.85% and 89.35%, respectively. We release our open-source code at https://github.com/andreibunea99/SegMate.

Problem

• medical image segmentation
• computational efficiency
• resource-constrained deployment
• multi-organ segmentation

Innovation

• asymmetric architecture
• attention mechanism
• multi-scale feature fusion
• slice-based positional conditioning
• multi-task optimization