Accurate Intervertebral Disc Segmentation Approach Based on Deep Learning

Area of Science:

• Medical Imaging
• Deep Learning
• Radiology

Background:

• Automated segmentation of medical images is common.
• Differentiating specific structures among similar targets remains challenging.
• Accurate intervertebral disc segmentation is crucial for spinal analysis.

Purpose of the Study:

• To develop an accurate method for segmenting specific intervertebral discs from lateral spine MRI scans.
• To address the challenge of distinguishing individual discs within a group of similar anatomical structures.

Main Methods:

• Utilized MultiResUNet models and saliency maps for intervertebral disc segmentation.
• Employed a sub-image cropping method based on saliency maps for initial disc isolation.
• Applied data augmentation using distance maps for segmenting remaining target discs.

Main Results:

• Achieved a segmentation accuracy of approximately 98% for specific intervertebral discs.
• Demonstrated significant enhancement in segmentation performance compared to existing methods.
• Successfully segmented target discs even when they were closely similar to adjacent discs.

Conclusions:

• The proposed MultiResUNet and saliency map approach effectively segments specific intervertebral discs in MRI.
• This method offers a robust solution for challenging segmentation tasks in medical imaging.
• The technique shows promise for improving diagnostic accuracy in spinal imaging analysis.