Feasibility of deep learning algorithm in diagnosing lumbar central canal stenosis using abdominal CT

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Lumbar central canal stenosis (LCCS) is a common condition requiring accurate diagnosis.
• Current diagnostic methods may involve invasive procedures or limited imaging modalities.
• Deep learning offers potential for automated and efficient LCCS diagnosis.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for diagnosing LCCS.
• To assess the algorithm's performance using both abdominal CT (ACT) and lumbar spine CT (LCT) images.
• To compare the diagnostic accuracy of ACT and LCT for LCCS detection via AI.

Main Methods:

• A retrospective study of 109 patients with available LCT and ACT scans.
• Manual segmentation of the dural sac on CT images to define normal and stenosed criteria.
• Development of a U-Net based deep learning model for automated dural sac segmentation and LCCS classification.

Main Results:

• The deep learning algorithm achieved high segmentation performance (DSC: 0.85, ICC: 0.82).
• The algorithm demonstrated strong agreement between ACT and LCT (ICC: 0.89).
• Overall diagnostic accuracy for LCCS was 84%, with ACT (85%) showing slightly higher accuracy than LCT (83%).

Conclusions:

• A deep learning algorithm can automatically diagnose LCCS using both LCT and ACT.
• Abdominal CT (ACT) provides diagnostic performance for LCCS comparable to lumbar spine CT (LCT).
• This AI approach may enhance the efficiency and accessibility of LCCS diagnosis.