Related Experiment Video
Updated: Jan 10, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Sustainable Ultralightweight U-Net-Based Architecture for Myocardium Segmentation
Jakub Filarecki1, Dorota Mockiewicz1, Agata Giełczyk1
1Faculty of Telecommunications, Computer Science and Electrical Engineering, Bydgoszcz University of Science and Technology, 85-796 Bydgoszcz, Poland.
Journal of Clinical Medicine
|November 27, 2025
Summary
A new lightweight AI model offers efficient cardiac muscle segmentation comparable to U-Net but with significantly fewer parameters. This sustainable approach aids radiologists in diagnosis and treatment planning.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Biomedical Engineering
Background:
- Medical image segmentation is crucial for diagnosis and treatment.
- U-Net is the standard but computationally intensive.
- Challenges exist in deploying large AI models in clinical settings.
Purpose of the Study:
- To develop a lightweight AI architecture for myocardium segmentation.
- To address the computational demands of current segmentation models.
- To improve the efficiency and sustainability of AI in medical imaging.
Main Methods:
- Utilized real-world MRI data from hospital patients.
- Proposed a novel, lightweight neural network architecture.
- Tailored the architecture specifically for cardiac muscle segmentation.
Main Results:
- Achieved segmentation performance comparable to state-of-the-art methods (IoU=0.7889, Dice=0.8780).
- The proposed model uses only 263k parameters and 6.24 G FLOPs.
- Demonstrated significant reduction in model complexity and computational cost.
Conclusions:
- The lightweight architecture supports radiologists in enhancing diagnostic accuracy.
- The approach is efficient, fast, and suitable for sustainable AI development.
- Offers a promising, less complex alternative to existing state-of-the-art methods.
