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

The paper by Jeon et al. (2024) explores the feasibility of using a deep learning algorithm to diagnose lumbar central canal stenosis (LCCS) through abdominal and lumbar CT scans. The study used a U-Net architecture to automatically segment the dural sac and classify stenosis based on a cross-sectional area threshold. The model was trained on 109 patients’ CT scans and achieved an accuracy of 84% in detecting LCCS, with better results using abdominal CT. The algorithm demonstrated a Dice similarity coefficient of 0.85, showing strong segmentation performance compared to manual radiologist measurements. The key contribution is the potential for abdominal CT scans, typically performed for other reasons, to detect LCCS, reducing the need for costly and invasive diagnostic methods. Limitations include the small dataset and absence of MRI comparison. Future work should focus on expanding the dataset and validating the algorithm in clinical practice. Coreline Soft participated in this research as an author.