[Development and validation of an automatic diagnostic tool for lumbar stability based on deep learning]

Area of Science:

• Spine Surgery
• Medical Imaging Analysis
• Deep Learning Applications in Healthcare

Context:

• Lumbar spine instability and spondylolisthesis are significant clinical concerns.
• Manual assessment of X-ray films for these conditions can be time-consuming and subject to inter-observer variability.
• Developing automated diagnostic tools is crucial for improving efficiency and consistency in clinical practice.

Purpose:

• To develop and validate a deep learning-based automatic diagnostic tool for lumbar spine stability.
• To assess the accuracy of the proposed Swin-PGNet model in identifying key lumbar vertebral points, calculating Cobb angles, and measuring lumbar sliding distance.
• To compare the diagnostic performance of Swin-PGNet against orthopedic surgeons for lumbar instability and spondylolisthesis.

Summary:

• A novel neural network, Swin-PGNet, was trained on 306 lumbar X-ray films to automatically locate key points and measure parameters related to spine stability.
• Swin-PGNet achieved comparable accuracy to orthopedic surgeons in key point localization, Cobb angle measurement, and lumbar sliding distance assessment.
• The tool demonstrated higher accuracy (84.0%) than surgeons (75.3%) in diagnosing lumbar instability, with similar performance in spondylolisthesis detection.

Impact:

• The developed deep learning tool provides an accurate and convenient method for automatic identification of lumbar instability and spondylolisthesis.
• This technology has the potential to significantly assist clinicians in making faster and more reliable diagnoses.
• Improved diagnostic accuracy can lead to better patient management and treatment outcomes for lumbar spine conditions.