Amanda J. L. Cheng

Purpose To compare deep learning models of different architecture for automated lumbar spinal stenosis classification on MRI and benchmark their performance against radiologists and orthopedists. Methods Lumbar spine MRI studies from Sep-2015 to Sep-2019 were retrospectively obtained. Exclusion criteria included previous spinal instrumentation, suboptimal image quality, post-gadolinium studies, and severe scoliosis. Axial T2-weighted and sagittal T1-weighted images were used. Studies were split into training/validation and test sets. An external test set of 100 studies was used. Training data were labelled by 4 radiologists using predefined gradings. Two models, CNN-based and transformer-based, were developed. Consensus labelling by two expert spine radiologists served as the reference standard. Test sets were labelled by 8 participants (2 general radiologists, 2 radiologists-in-training, 2 orthopedists, 2 orthopedists-in-training). Detection recall (%), interrater agreement (Gwet κ), sensitivity, and specificity were evaluated. Results 564 MRI lumbar spines were included (mean age = 52 ± 19[SD]; 302 women), with 464(82%) and 100(18%) for training/validation and internal testing, respectively. Both models showed high recall for all regions of interest (> 94%), similar to participants. Dichotomous classification (normal/mild vs. moderate/severe) by the CNN model, transformer model, and participants showed respective kappas for central canal 0.99/0.99/0.97–0.98, lateral recesses 0.98/0.94/0.81–0.94, and neural foramina 0.98/0.95/0.91–0.95 on internal testing (p < 0.001); for central canal 0.99/0.97/0.92–0.97, lateral recess 0.97/0.90/0.61–0.91, and neural foramina 0.99/0.94/0.87–0.93 on external testing (p < 0.001). Conclusion The CNN model showed superior performance, and the transformer model showed similar to superior performance compared to clinicians for classifying lumbar spinal stenosis. These models could assist clinicians in report generation, surgical planning and education.

Purpose To develop a deep learning (DL) model for epidural spinal cord compression (ESCC) on CT, which will aid earlier ESCC diagnosis for less experienced clinicians. Methods We retrospectively collected CT and MRI data from adult patients with suspected ESCC at a tertiary referral institute from 2007 till 2020. A total of 183 patients were used for training/validation of the DL model. A separate test set of 40 patients was used for DL model evaluation and comprised 60 staging CT and matched MRI scans performed with an interval of up to 2 months. DL model performance was compared to eight readers: one musculoskeletal radiologist, two body radiologists, one spine surgeon, and four trainee spine surgeons. Diagnostic performance was evaluated using inter-rater agreement, sensitivity, specificity and AUC. Results Overall, 3115 axial CT slices were assessed. The DL model showed high kappa of 0.872 for normal, low and high-grade ESCC (trichotomous), which was superior compared to a body radiologist (R4, κ = 0.667) and all four trainee spine surgeons (κ range = 0.625–0.838)(all p < 0.001). In addition, for dichotomous normal versus any grade of ESCC detection, the DL model showed high kappa (κ = 0.879), sensitivity (91.82), specificity (92.01) and AUC (0.919), with the latter AUC superior to all readers (AUC range = 0.732–0.859, all p < 0.001). Conclusion A deep learning model for the objective assessment of ESCC on CT had comparable or superior performance to radiologists and spine surgeons. Earlier diagnosis of ESCC on CT could reduce treatment delays, which are associated with poor outcomes, increased costs, and reduced survival.

Radiologists assisted by deep learning for interpretation of lumbar spinal stenosis on MRI scans showed a marked reduction in reporting time and superior or equivalent interobserver agreement for a... Background Lumbar spine MRI studies are widely used for back pain assessment. Interpretation involves grading lumbar spinal stenosis, which is repetitive and time consuming. Deep learning (DL) coul...

Background: Metastatic epidural spinal cord compression (MESCC) is a disastrous complication of advanced malignancy. Deep learning (DL) models for automatic MESCC classification on staging CT were developed to aid earlier diagnosis. Methods: This retrospective study included 444 CT staging studies from 185 patients with suspected MESCC who underwent MRI spine studies within 60 days of the CT studies. The DL model training/validation dataset consisted of 316/358 (88%) and the test set of 42/358 (12%) CT studies. Training/validation and test datasets were labeled in consensus by two subspecialized radiologists (6 and 11-years-experience) using the MRI studies as the reference standard. Test sets were labeled by the developed DL models and four radiologists (2–7 years of experience) for comparison. Results: DL models showed almost-perfect interobserver agreement for classification of CT spine images into normal, low, and high-grade MESCC, with kappas ranging from 0.873–0.911 (p < 0.001). The DL models (lowest κ = 0.873, 95% CI 0.858–0.887) also showed superior interobserver agreement compared to two of the four radiologists for three-class classification, including a specialist (κ = 0.820, 95% CI 0.803–0.837) and general radiologist (κ = 0.726, 95% CI 0.706–0.747), both p < 0.001. Conclusion: DL models for the MESCC classification on a CT showed comparable to superior interobserver agreement to radiologists and could be used to aid earlier diagnosis.