Related Experiment Video
Updated: Jan 7, 2026
05:49
Author Spotlight: Deciphering Electrical Networks Behind Complex Brain Activities and Disorders
Published on: November 1, 2024
1.2K
MPG-SwinUMamba: High-Precision Segmentation and Automated Measurement of Eye Muscle Area in Live Sheep Based on Deep
Zhou Zhang1,2,3, Yaojing Yue4, Fuzhong Li2
1College of Agricultural Engineering, Shanxi Agricultural University, Jinzhong 030801, China.
Animals : an Open Access Journal From MDPI
|December 30, 2025
Summary
A new deep learning model, MPG-SwinUMamba, accurately measures the eye muscle area (EMA) in live sheep using ultrasound images. This non-invasive method enhances genetic breeding and production management in the meat sheep industry.
Area of Science:
- Veterinary Medicine
- Animal Science
- Biomedical Engineering
Background:
- Accurate eye muscle area (EMA) assessment is vital for genetic breeding and production management in the meat sheep industry.
- Existing automated methods struggle with B-mode ultrasound image quality, including low contrast and noise, limiting segmentation accuracy.
- Developing robust automated segmentation is essential for efficient and objective evaluation of carcass performance in live sheep.
Purpose of the Study:
- To introduce MPG-SwinUMamba, a novel deep learning segmentation network designed to overcome the limitations of current automated EMA assessment in live sheep.
- To improve the accuracy and reliability of EMA measurements from B-mode ultrasound images.
- To provide a non-invasive, efficient, and objective tool for evaluating carcass performance and enhancing breeding efficiency.
Main Methods:
- Developed MPG-SwinUMamba, a deep learning network combining a state-space model with a U-Net architecture.
- Integrated edge-enhancement multi-scale attention (MSEE) and pyramid attention refinement (PARM) modules to enhance boundary detection and global context capture.
- Utilized a global context aggregation decoder (GCAD) for precise segmentation mask reconstruction and automated EMA measurement.
Main Results:
- MPG-SwinUMamba outperformed 12 other segmentation models, achieving an intersection-over-union (IoU) of 91.62% and a Dice similarity coefficient (DSC) of 95.54%.
- Automated EMA measurements demonstrated strong agreement with expert manual assessments, showing a correlation coefficient (r) of 0.9637.
- The mean absolute percentage error (MAPE) for automated measurements was only 4.05%, indicating high accuracy.
Conclusions:
- MPG-SwinUMamba provides a highly accurate and reliable method for automated EMA segmentation and measurement in live sheep using ultrasound.
- The non-invasive approach has significant potential to reduce measurement costs and improve breeding efficiency in the meat sheep industry.
- This deep learning model offers a valuable tool for objective evaluation of carcass performance, supporting genetic selection and production management.
