Related Experiment Video
Updated: Jan 13, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
732
A Lesion-Aware Patch Sampling Approach with EfficientNet3D-UNet for Robust Multiple Sclerosis Lesion Segmentation
Hind Almaaz1, Samia Dardouri1,2
1College of Computing and Information Technology, Shaqra University, Shaqra 11911, Saudi Arabia.
Journal of Imaging
|October 28, 2025
Summary
A new deep learning model, Efficient-Net3D-UNet, significantly improves automated segmentation of multiple sclerosis (MS) lesions in 3D MRI scans. This advancement offers a more accurate and efficient tool for clinical diagnosis and monitoring.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Accurate segmentation of multiple sclerosis (MS) lesions in 3D MRI is critical for patient care.
- Challenges include lesion subtlety, heterogeneity, and annotation difficulties.
Purpose of the Study:
- To develop an improved deep learning framework for automated MS lesion segmentation.
- To enhance volumetric segmentation performance across multi-modal MRI sequences.
Main Methods:
- Proposed Efficient-Net3D-UNet, integrating MBConv3D blocks and lesion-aware patch sampling.
- Evaluated against a conventional 3D U-Net baseline.
- Utilized Dice similarity coefficient, precision, recall, accuracy, and specificity for assessment.
Main Results:
- EfficientNet3D-UNet achieved a Dice score of 48.39%, outperforming the baseline 3D U-Net (31.28%).
- EfficientNet3D-UNet demonstrated higher precision (49.76%) and recall (55.41%).
- The proposed model showed faster convergence and reduced overfitting.
Conclusions:
- Efficient-Net3D-UNet presents a robust and computationally efficient solution for MS lesion segmentation.
- The model shows promise for real-world clinical applications in automated diagnosis and monitoring.
