Development of a Deep-Learning Model for Diagnosing Lumbar Spinal Stenosis Based on CT Images

Area of Science:

• Radiology
• Artificial Intelligence
• Spine Surgery

Background:

• Computed tomography (CT) is a cost-effective and portable imaging modality for diagnosing lumbar spinal stenosis, offering wider coverage than MRI.
• Deep learning (DL) models can enhance CT scan accuracy, potentially reducing missed diagnoses and misdiagnoses in clinical settings.

Purpose of the Study:

• To develop an initial deep-learning (DL) model utilizing CT scans for diagnosing lumbar spinal stenosis.
• To evaluate the diagnostic accuracy of the DL model in comparison to specialist physicians.

Main Methods:

• A retrospective study involving axial lumbar spine CT scans from March 2022 to September 2023.
• Development of a DL model, including region of interest detection and a convolutional neural network classifier, trained on a labeled dataset.
• Performance evaluation of the DL model on a control set, comparing its accuracy against that of spine specialists.

Main Results:

• The DL model achieved 88% accuracy in grading central stenosis (DL Model Version 1) and 75% accuracy in grading lateral recess stenosis (DL Model Version 1).
• DL Model Version 2 showed slightly lower accuracies of 83% for central stenosis and 71% for lateral recess stenosis.

Conclusions:

• The preliminary DL system demonstrates comparable accuracy to experienced physicians in assessing lumbar spinal stenosis severity via CT.
• The developed DL model holds significant potential for future development and clinical integration in diagnosing lumbar spinal stenosis.