[2602.23903] SegMate: Asymmetric Attention-Based Lightweight Architecture for Efficient Multi-Organ Segmentation

Computer Science > Computer Vision and Pattern Recognition arXiv:2602.23903 (cs) [Submitted on 27 Feb 2026] Title:SegMate: Asymmetric Attention-Based Lightweight Architecture for Efficient Multi-Organ Segmentation Authors:Andrei-Alexandru Bunea, Dan-Matei Popovici, Radu Tudor Ionescu View a PDF of the paper titled SegMate: Asymmetric Attention-Based Lightweight Architecture for Efficient Multi-Organ Segmentation, by Andrei-Alexandru Bunea and 2 other authors View PDF HTML (experimental) 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 s...