Related Experiment Video
Updated: Jun 18, 2026
06:45
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
Balancing accuracy and efficiency in lumbar spine image segmentation using multi-scale attention and residual
Vijaya Bhaskar Sadu1, Bharathi Subramaniam2, Rukmani Devi Sethuraman3
1Department of Mechanical Engineering, Jawaharlal Nehru Technological University, Kakinada, Andhra Pradesh, 533003, India.
Scientific Reports
|June 16, 2026
Summary
A novel deep learning model, MRAS-FPN, significantly improves Lumbar Spine Segmentation (LSS) in MRI scans. It achieves state-of-the-art accuracy and efficiency, aiding in diagnosis and treatment planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Accurate Lumbar Spine Segmentation (LSS) in MRI is crucial for clinical applications.
- Existing methods often struggle with a precision-efficiency trade-off and modeling complex anatomical details.
Purpose of the Study:
- To introduce a novel Deep Learning (DL) model, Multiscale Residual and Adaptive Spatial-Channel Feature Pyramid Network (MRAS-FPN), for enhanced LSS.
- To improve segmentation accuracy and computational efficiency in lumbar spine MRI analysis.
Main Methods:
- Developed MRAS-FPN integrating Multi-Scale Attention (MS) and Residual Learning (RL).
- Incorporated Multiscale Residual Attention Blocks (MRAB) with Directionally Dilated Convolution Modules (DDCM) and Adaptive Spatial-Channel Attention Blocks (ASCAB).
- Trained and validated on the MRSpineSeg dataset (215 T2-weighted lumbar MRI volumes).
Main Results:
- MRAS-FPN achieved a Dice Similarity Coefficient (DSC) of 87.8% and mIoU of 78.2%, outperforming 3D U-Net.
- Demonstrated significant improvements in boundary accuracy (HD95: 3.18 mm, ASSD: 1.32 mm).
- Showed stable segmentation quality across vertebrae (T9-S1) and intervertebral discs, with notable gains in complex regions.
Conclusions:
- MRAS-FPN offers a superior balance of segmentation fidelity and resource efficiency for LSS.
- The model's performance and ability to capture intricate details make it suitable for clinical applications like surgical planning and diagnostic support.
- MRAS-FPN represents a significant advancement in automated lumbar spine MRI analysis.
