Localization and Edge-Based Segmentation of Lumbar Spine Vertebrae to Identify the Deformities Using Deep Learning...

Area of Science:

• Medical Imaging
• Computer Vision
• Spine Surgery

Background:

• Accurate lumbar spine analysis is crucial for diagnosing deformities and fractures.
• Manual diagnosis methods are time-consuming and prone to inter-observer variability.
• Automated medical image analysis can enhance diagnostic efficiency and consistency.

Purpose of the Study:

• To develop and evaluate automated methods for lumbar spine localization and segmentation.
• To improve the accuracy and reliability of diagnosing lumbar spine deformities, such as lumbar lordosis.
• To provide clinicians with tools for confident disease severity grading.

Main Methods:

• Lumbar spine localization using YOLOv5 object detection, achieving a mean average precision (mAP) of 0.975.
• Segmentation of vertebrae and their edges using HED U-Net on YOLOv5-cropped images.
• Calculation of lumbar lordotic angles (LLAs) and lumbosacral angles (LSAs) using Harris corner detection.

Main Results:

• YOLOv5 demonstrated high performance in lumbar vertebrae localization.
• The combined approach achieved 74.5% accuracy in diagnosing lumbar lordosis by correlating angles with region area.
• Harris corner detection yielded minimal mean errors of 0.29° for LLAs and 0.38° for LSAs.

Conclusions:

• Automated YOLOv5 and HED U-Net methods provide accurate lumbar spine localization and segmentation.
• The developed techniques offer a reliable approach for diagnosing lumbar deformities and grading disease severity.
• This study highlights the potential of AI in improving spinal diagnostics compared to traditional methods.