Deep learning-based detection of lumbar spinal canal stenosis using convolutional neural networks

Lumbar spinal canal stenosis (LSCS) is the most common spinal degenerative disorder in elderly people and usually first seen by primary care physicians or orthopedic surgeons who are not spine surgery specialists. Magnetic resonance imaging (MRI) is useful in the diagnosis of LSCS, but the equipment is often not available or difficult to read. LSCS patients with progressive neurologic deficits have difficulty with recovery if surgical treatment is delayed. So, early diagnosis and determination of appropriate surgical indications are crucial in the treatment of LSCS. Convolutional neural networks (CNNs), a type of deep learning, offers significant advantages for image recognition and classification, and work well with radiographs, which can be easily taken at any facility.

Our purpose was to develop an algorithm to diagnose the presence or absence of LSCS requiring surgery from plain radiographs using CNNs.

Retrospective analysis of consecutive, nonrandomized series of patients at a single institution.

Data of 150 patients who underwent surgery for LSCS, including degenerative spondylolisthesis, at a single institution from January 2022 to August 2022, were collected. Additionally, 25 patients who underwent surgery at 2 other hospitals were included for extra external validation.

In annotation 1, the area under the curve (AUC) computed from the receiver operating characteristic (ROC) curve, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, positive likelihood ratio (PLR), and negative likelihood ratio (NLR) were calculated. In annotation 2, correlation coefficients were used.

Four intervertebral levels from L1/2 to L4/5 were extracted as region of interest from lateral plain lumbar spine radiographs totaling 600 images were obtained. Based on the date of surgery, 500 images derived from the first 125 cases were used for internal validation, and 100 images from the subsequent 25 cases used for external validation. Additionally, 100 images from other hospitals were used for extra external validation. In annotation 1, binary classification of operative and nonoperative levels was used, and in annotation 2, the spinal canal area measured on axial MRI was labeled as the output layer. For internal validation, the 500 images were divided into each 5 dataset on per-patient basis and 5-fold cross-validation was performed. Five trained models were registered in the external validation prediction performance. Grad-CAM was used to visualize area with the high features extracted by CNNs.

In internal validation, the AUC and accuracy for annotation 1 ranged between 0.85-0.89 and 79-83%, respectively, and the correlation coefficients for annotation 2 ranged between 0.53 and 0.64 (all p<.01). In external validation, the AUC and accuracy for annotation 1 were 0.90 and 82%, respectively, and the correlation coefficient for annotation 2 was 0.69, using 5 trained CNN models. In the extra external validation, the AUC and accuracy for annotation 1 were 0.89 and 84%, respectively, and the correlation coefficient for annotation 2 was 0.56. Grad-CAM showed high feature density in the intervertebral joints and posterior intervertebral discs.

This technology automatically detects LSCS from plain lumbar spine radiographs, making it possible for medical facilities without MRI or nonspecialists to diagnose LSCS, suggesting the possibility of eliminating delays in the diagnosis and treatment of LSCS that require early treatment.