Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LSU-Net: Lightweight Automatic Organs Segmentation Network For Medical Images

Yujie Ding, Shenghua Teng, Zuoyong Li, Xiao Chen

TL;DR

LSU-Net addresses the need for efficient abdominal organ segmentation by combining lightweight convolutional blocks with a tokenized shift mechanism and multi-scale supervision. The two novel modules, Light Conv Block and Tokenized Shift Block, enable low-parameter feature extraction and enhanced spatial–channel representation, while MDL and AWL provide stable, adaptive multi-scale training. The approach achieves competitive accuracy on the UWMGI and MSD Colon datasets with a fraction of the parameters and computation of many rivals, and ablation confirms the importance of each component. This facilitates practical deployment in resource-constrained clinical environments for tasks like lesion quantification and progression monitoring.

Abstract

UNet and its variants have widespread applications in medical image segmentation. However, the substantial number of parameters and computational complexity of these models make them less suitable for use in clinical settings with limited computational resources. To address this limitation, we propose a novel Lightweight Shift U-Net (LSU-Net). We integrate the Light Conv Block and the Tokenized Shift Block in a lightweight manner, combining them with a dynamic weight multi-loss design for efficient dynamic weight allocation. The Light Conv Block effectively captures features with a low parameter count by combining standard convolutions with depthwise separable convolutions. The Tokenized Shift Block optimizes feature representation by shifting and capturing deep features through a combination of the Spatial Shift Block and depthwise separable convolutions. Dynamic adjustment of the loss weights at each layer approaches the optimal solution and enhances training stability. We validated LSU-Net on the UWMGI and MSD Colon datasets, and experimental results demonstrate that LSU-Net outperforms most state-of-the-art segmentation architectures.

LSU-Net: Lightweight Automatic Organs Segmentation Network For Medical Images

TL;DR

LSU-Net addresses the need for efficient abdominal organ segmentation by combining lightweight convolutional blocks with a tokenized shift mechanism and multi-scale supervision. The two novel modules, Light Conv Block and Tokenized Shift Block, enable low-parameter feature extraction and enhanced spatial–channel representation, while MDL and AWL provide stable, adaptive multi-scale training. The approach achieves competitive accuracy on the UWMGI and MSD Colon datasets with a fraction of the parameters and computation of many rivals, and ablation confirms the importance of each component. This facilitates practical deployment in resource-constrained clinical environments for tasks like lesion quantification and progression monitoring.

Abstract

UNet and its variants have widespread applications in medical image segmentation. However, the substantial number of parameters and computational complexity of these models make them less suitable for use in clinical settings with limited computational resources. To address this limitation, we propose a novel Lightweight Shift U-Net (LSU-Net). We integrate the Light Conv Block and the Tokenized Shift Block in a lightweight manner, combining them with a dynamic weight multi-loss design for efficient dynamic weight allocation. The Light Conv Block effectively captures features with a low parameter count by combining standard convolutions with depthwise separable convolutions. The Tokenized Shift Block optimizes feature representation by shifting and capturing deep features through a combination of the Spatial Shift Block and depthwise separable convolutions. Dynamic adjustment of the loss weights at each layer approaches the optimal solution and enhances training stability. We validated LSU-Net on the UWMGI and MSD Colon datasets, and experimental results demonstrate that LSU-Net outperforms most state-of-the-art segmentation architectures.

Paper Structure

This paper contains 4 sections, 13 equations, 3 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. METHOD
  3. EXPERIMENTS
  4. CONCLUSION

Figures (3)

  • Figure 1: The overview of the Lightweight Shift U-Net(LSU-Net). The downsampling in the encoder is achieved by setting the convolution stride to 2. Deep Supervision is used at each layer. $l_{0}$ to $l_{5}$ means loss function in each level. The upper right of the figure shows the Tokenized Shift Block, while the lower right of the figure shows the Light Conv Block.
  • Figure 2: The illustration of spatial shifts in the Spatial Shift Block. Panels (a) and (b) show two different shift methods applied to the height and width, differing in the order of the shifts.
  • Figure 3: Visualization of segmentation results on the UWMGI and MSD Colon Cancer datasets. The above visualization shows MSD Colon Cancer results, while the bottom shows UWMGI results. From left to right: origin image, ground truth and the results of UNeXtUNext, Res-UNetResUNet, UNetUNet, UNet++UNet++, Attention-UNetAttention-UNet, and LSU-Net.