LightMedSeg: Lightweight 3D Medical Image Segmentation with Learned Spatial Anchors

Kavyansh Tyagi; Vishwas Rathi; Puneet Goyal

cs.LGMar 7, 2026

LightMedSeg: Lightweight 3D Medical Image Segmentation with Learned Spatial Anchors

Authors: Kavyansh Tyagi, Vishwas Rathi, Puneet Goyal

Organizations: National Institute of Technology Kurukshetra, India · Indian Institute of Technology Ropar, India

Abstract

Accurate and efficient 3D medical image segmentation is essential for clinical AI, where models must remain reliable under stringent memory, latency, and data availability constraints. Transformer-based methods achieve strong accuracy but suffer from excessive parameters, high FLOPs, and limited generalization. We propose LightMedSeg, a modular UNet-style segmentation architecture that integrates anatomical priors with adaptive context modeling. Anchor-conditioned FiLM modulation enables anatomy-aware feature calibration, while a local structural prior module and texture-aware routing dynamically allocate representational capacity to boundary-rich regions. Computational redundancy is minimized through ghost and depthwise convolutions, and multi-scale features are adaptively fused via a learned skip router with anchor-relative spatial position bias. Despite requiring only 0.48M parameters and 14.64~GFLOPs, LightMedSeg achieves segmentation accuracy within a few Dice points of heavy transformer baselines. Therefore, LightMedSeg is a deployable and data-efficient solution for 3D medical image segmentation. Code will be released publicly upon acceptance.

Figures & tables

Figure 1: Architecture of LightMedSeg . The input volume is first embedded by a stride-2 GhostConv3D stem. The Global Anchor Detector and the Local Structural Prior Module (LSPM) extract spatial anchors

\mathbf{S}

and a texture routing map

\mathbf{T}

from the stem features. These priors guide (i) adaptive feature mixing before the encoder, (ii) anchor-conditioned, texture-routed processing at every encoder stage, (iii) multi-scale learned skip fusion, and (iv) an adaptive decoder with content-tied spatial position bias. A final ConvTranspose3D restores voxel-wise class logits at the original resolution.

Methods	Param	Avg %	Whole Tumor ↑	Enhancing Tumor ↑	Tumor Core ↑
nnFormer [ 31 ]	150.5	86.4	91.3	81.8	86.0
Segformer3D [ 20 ]	4.5	82.1	89.9	74.2	82.2
UNETR [ 10 ]	92.49	71.1	78.9	58.5	76.1
TransBTS [ 23 ]	–	69.6	77.9	57.4	73.5
CoTr [ 26 ]	41.9	68.3	74.6	55.7	74.8
CoTr w/o CNN Encoder [ 26 ]	–	64.4	71.2	52.3	69.8

Table 1: Comparison of various methods on the BraTS dataset using average Dice score. Bold values represent the best performance, while underlined values indicate the second best results. Here, the number of parameters (denoted as ’params’) is reported in millions.

Panel of Figure 2 — Figure 2: Visual comparison of original MRI, ground truth, and predicted segmentation for four representative BraTS cases.

Panel of Figure 3 — Figure 3: Visual comparison of original MRI, ground truth, and predicted segmentation for four ACDC cases.

Method	Params	Avg%↑	RV	Myo	LV
nnFormer [ 31 ]	150.5	92.06	90.94	89.58	95.65
Segformer3D [ 20 ]	4.5	90.96	88.50	88.86	95.53
LeViT-UNet [ 27 ]	52.17	90.32	89.55	87.64	93.76
SwinUNet [ 3 ]	–	90.00	88.55	85.62	95.83
TransUNet [ 4 ]	96.07	89.71	88.86	85.54	95.73
UNETR [ 10 ]	92.49	88.61	85.29	86.52	94.02

Table 2: Comparison on ACDC dataset (avg. Dice score). Params in millions. Bold : best, underline : second best.

Architecture	Params (M)	FLOPs (G)	Mem (MB)
nnFormer [ 31 ]	150.5	213.4	12902
UNETR [ 10 ]	92.49	75.76	3379
SwinUNETR [ 9 ]	62.83	384.2	20173
UNETR++ [ 22 ]	42.96	70.1	2458
SegFormer3D [ 20 ]	4.51	17.5	–
LightMedSeg	0.48	14.64	1174

Table 3: Comparison of LightMedSeg with state-of-the-art methods in terms of model size, computational complexity, and peak memory. Measured on a BraTS-sized input (

1\!\times\!4\!\times\!128^{3}

); – indicates unavailable data.

Configuration	Avg Dice
Full Model	83.43
w/o LSPM	80.50 ( $-2.93$ )
w/o Anchors	81.93 ( $-1.50$ )
w/o Skip Router	82.27 ( $-1.16$ )
w/o GhostConv3D	84.81 ( $+1.38$ )
$K{=}8$ (default)	83.43 (0.48M)

Table 4: Component ablation on BraTS (Avg. Dice).

K

denotes the number of anatomical anchors.

Explore similar work

RefineFormer3D: Efficient 3D Medical Image Segmentation via Adaptive Multi-Scale Transformer with Cross Attention Fusion
Feb 18, 2026Kavyansh Tyagi, Vishwas Rathi, Puneet Goyal3D Medical Image Segmentation 3D ViTs
National Institute of Technology Kurukshetra Department of Electronics and Communication Engineering · National Institute of Technology Kurukshetra Department of Computer Engineering · Indian Institute of Technology Ropar Department of Computer Science and Engineering
Harnessing Lightweight Transformer with Contextual Synergic Enhancement for Efficient 3D Medical Image Segmentation
Mar 24, 2026Xinyu Liu, Zhen Chen, Wuyang Li +23D Medical Image Segmentation 3D ViTs
Department of Electronic Engineering, The Chinese University of Hong Kong, Hong Kong SAR · Centre for Artificial Intelligence and Robotics (CAIR), Hong Kong Institute of Science & Innovation, Chinese Academy of Sciences, Hong Kong SAR
SegMaFormer: A Hybrid State-Space and Transformer Model for Efficient Segmentation
Mar 23, 2026Duy D. Nguyen, Phat T. Tran-Truong3D Medical Image Segmentation 3D ViTs
Department of Software Engineering, Faculty of Computer Science and Engineering, Ho Chi Minh City University of Technology (HCMUT), 268 Ly Thuong Kiet Street, Dien Hong Ward, Ho Chi Minh City, Vietnam · Vietnam National University Ho Chi Minh City, Linh Trung Ward, Ho Chi Minh City, Vietnam

Feb 18, 2026J/K move · Enter open · S save

RefineFormer3D: Efficient 3D Medical Image Segmentation via Adaptive Multi-Scale Transformer with Cross Attention Fusion

Kavyansh Tyagi, Vishwas Rathi, Puneet Goyal

National Institute of Technology Kurukshetra Department of Electronics and Communication Engineering · National Institute of Technology Kurukshetra Department of Computer Engineering · Indian Institute of Technology Ropar Department of Computer Science and Engineering

3D Medical Image Segmentation 3D ViTs Efficient ViTs Hybrid CNN-Transformer Architectures

Open paper AskPDF arXiv

Accurate and computationally efficient 3D medical image segmentation remains a critical challenge in clinical workflows. Transformer-based architectures often demonstrate superior global contextual modeling but at the expense of excessive parameter counts and memory demands, restricting their clinical deployment. We propose RefineFormer3D, a lightweight hierarchical transformer architecture that balances segmentation accuracy and computational efficiency for volumetric medical imaging. The architecture integrates three key components: (i) GhostConv3D-based patch embedding for efficient feature extraction with minimal redundancy, (ii) MixFFN3D module with low-rank projections and depthwise convolutions for parameter-efficient feature extraction, and (iii) a cross-attention fusion decoder enabling adaptive multi-scale skip connection integration. RefineFormer3D contains only 2.94M parameters, substantially fewer than contemporary transformer-based methods. Extensive experiments on ACDC and BraTS benchmarks demonstrate that RefineFormer3D achieves 93.44% and 85.9% average Dice scores respectively, outperforming or matching state-of-the-art methods while requiring significantly fewer parameters. Furthermore, the model achieves fast inference (8.35 ms per volume on GPU) with low memory requirements, supporting deployment in resource-constrained clinical environments. These results establish RefineFormer3D as an effective and scalable solution for practical 3D medical image segmentation.

Abstract

Explore similar work

RefineFormer3D: Efficient 3D Medical Image Segmentation via Adaptive Multi-Scale Transformer with Cross Attention Fusion

Harnessing Lightweight Transformer with Contextual Synergic Enhancement for Efficient 3D Medical Image Segmentation

SegMaFormer: A Hybrid State-Space and Transformer Model for Efficient Segmentation