牧 聡

STUDY DESIGN: A retrospective analysis. OBJECTIVE: This research sought to develop a predictive model for surgical outcomes in patients with cervical ossification of the posterior longitudinal ligament (OPLL) using deep learning and machine learning (ML) techniques. SUMMARY OF BACKGROUND DATA: Determining surgical outcomes assists surgeons in communicating prognosis to patients and setting their expectations. Deep learning and ML are computational models that identify patterns from large datasets and make predictions. METHODS: Of the 482 patients, 288 patients were included in the analysis. A minimal clinically important difference (MCID) was defined as gain in Japanese Orthopaedic Association (JOA) score of 2.5 points or more. The predictive model for MCID achievement at 1 year post-surgery was constructed using patient background, clinical symptoms, and preoperative imaging features (x-ray, CT, MRI) analyzed via LightGBM and deep learning with RadImagenet. RESULTS: The median preoperative JOA score was 11.0 (IQR: 9.0-12.0), which significantly improved to 14.0 (IQR: 12.0-15.0) at 1 year after surgery (P < 0.001, Wilcoxon signed-rank test). The average improvement rate of the JOA score was 44.7%, and 60.1% of patients achieved the MCID. Our model exhibited an area under the receiver operating characteristic curve of 0.81 and the accuracy of 71.9% in predicting MCID at 1 year. Preoperative JOA score and certain preoperative imaging features were identified as the most significant factors in the predictive models. CONCLUSION: A predictive ML and deep learning model for surgical outcomes in OPLL patients is feasible, suggesting promising applications in spinal surgery. LEVEL OF EVIDENCE: 4.

In recent years, fracture image diagnosis using a convolutional neural network (CNN) has been reported. The purpose of the present study was to evaluate the ability of CNN to diagnose distal radius fractures (DRFs) using frontal and lateral wrist radiographs. We included 503 cases of DRF diagnosed by plain radiographs and 289 cases without fracture. We implemented the CNN model using Keras and Tensorflow. Frontal and lateral views of wrist radiographs were manually cropped and trained separately. Fine-tuning was performed using EfficientNets. The diagnostic ability of CNN was evaluated using 150 images with and without fractures from anteroposterior and lateral radiographs. The CNN model diagnosed DRF based on three views: frontal view, lateral view, and both frontal and lateral view. We determined the sensitivity, specificity, and accuracy of the CNN model, plotted a receiver operating characteristic (ROC) curve, and calculated the area under the ROC curve (AUC). We further compared performances between the CNN and three hand orthopedic surgeons. EfficientNet-B2 in the frontal view and EfficientNet-B4 in the lateral view showed highest accuracy on the validation dataset, and these models were used for combined views. The accuracy, sensitivity, and specificity of the CNN based on both anteroposterior and lateral radiographs were 99.3, 98.7, and 100, respectively. The accuracy of the CNN was equal to or better than that of three orthopedic surgeons. The AUC of the CNN on the combined views was 0.993. The CNN model exhibited high accuracy in the diagnosis of distal radius fracture with a plain radiograph.

STUDY DESIGN: Retrospective study of magnetic resonance imaging (MRI). OBJECTIVES: To assess the ability of a convolutional neural network (CNN) model to differentiate osteoporotic vertebral fractures (OVFs) and malignant vertebral compression fractures (MVFs) using short-TI inversion recovery (STIR) and T1-weighted images (T1WI) and to compare it to the performance of three spine surgeons. SUMMARY OF BACKGROUND DATA: Differentiating between OVFs and MVFs is crucial for appropriate clinical staging and treatment planning. However, an accurate diagnosis is sometimes difficult. Recently, CNN modeling-an artificial intelligence technique-has gained popularity in the radiology field. METHODS: We enrolled 50 patients with OVFs and 47 patients with MVFs who underwent thoracolumbar MRI. Sagittal STIR images and sagittal T1WI were used to train and validate the CNN models. To assess the performance of the CNN, the receiver operating characteristic (ROC) curve was plotted and the area under the curve (AUC) was calculated. We also compared the accuracy, sensitivity, and specificity of the diagnosis made by the CNN and three spine surgeons. RESULTS: The AUC of ROC curves of the CNN based on STIR images and T1WI were 0.967 and 0.984, respectively. The CNN model based on STIR images showed a performance of 93.8% accuracy, 92.5% sensitivity, and 94.9% specificity. On the other hand, the CNN model based on T1WI showed a performance of 96.4% accuracy, 98.1% sensitivity, and 94.9% specificity. The accuracy and specificity of the CNN using both STIR and T1WI were statistically equal to or better than that of three spine surgeons. There were no significant differences in sensitivity based on both STIR images and T1WI between the CNN and spine surgeons. CONCLUSIONS: We successfully differentiated OVFs and MVFs based on MRI with high accuracy using the CNN model, which was statistically equal or superior to that of the spine surgeons.Level of Evidence: 4.

STUDY DESIGN: A retrospective analysis of prospectively collected data. OBJECTIVE: This study aimed to create a prognostic model for surgical outcomes in patients with cervical ossification of the posterior longitudinal ligament (OPLL) using machine learning (ML). SUMMARY OF BACKGROUND DATA: Determining surgical outcomes helps surgeons provide prognostic information to patients and manage their expectations. ML is a mathematical model that finds patterns from a large sample of data and makes predictions outperforming traditional statistical methods. METHODS: Out of 478 patients, 397 and 370 patients had complete follow-up information at 1 and 2 year respectively and were included in the analysis. A minimal clinically important difference (MCID) was defined as an acquired Japanese Orthopaedic Association (JOA) score of 2.5 points or more, after which a ML model that predicts whether MCID can be achieved 1 and 2 years after surgery was created. Patient background, clinical symptoms, and imaging findings were used as variables for analysis. The ML model was created using LightGBM, XGBoost, random forest, and logistic regression, after which the accuracy and area under the receiver operating characteristic curve (AUC) were calculated. RESULTS: The mean JOA score was 10.3 preoperatively, 13.4 at 1 year after surgery, and 13.5 at 2 years after surgery. XGBoost showed the highest AUC (0.72) and high accuracy (67.8) for predicting MCID at 1 year, while random forest had the highest AUC (0.75) and accuracy (69.6) for predicting MCID at 2 years. Among the included features, total preoperative JOA score, duration of symptoms, body weight, sensory function of the lower extremity sub-score of the JOA, and age were identified as having the most significance in most of ML models. CONCLUSION: Constructing a prognostic ML model for surgical outcomes in patients with OPLL is feasible, suggesting the potential application of ML for predictive models of spinal surgery.Level of Evidence: 4.

PURPOSE: To demonstrate the feasibility of 3D multi-shot magnetic resonance imaging acquisitions for stimulus-evoked blood oxygenation level dependent (BOLD) functional magnetic resonance imaging (fMRI) in the human spinal cord in vivo. METHODS: Two fMRI studies were performed at 3T. The first study was a hypercapnic gas challenge where data were acquired from healthy volunteers using a multi-shot 3D fast field echo (FFE) sequence as well as single-shot multi-slice echo-planar imaging (EPI). In the second study, another cohort of healthy volunteers performed an upper extremity motor task while fMRI data were acquired using a 3D multi-shot acquisition. RESULTS: Both 2D-EPI and 3D-FFE were shown to be sensitive to BOLD signal changes in the cervical spinal cord, and had comparable contrast-to-noise ratios in gray matter. FFE exhibited much less signal drop-out and weaker geometric distortions compared to EPI. In the motor paradigm study, the mean number of active voxels was highest in the ventral gray matter horns ipsilateral to the side of the task and at the spinal level associated with innervation of finger extensors. CONCLUSIONS: Highly multi-shot acquisition sequences such as 3D-FFE are well suited for stimulus-evoked spinal cord BOLD fMRI.

BACKGROUND: According to most of the commonly used classification systems for subaxial spine injuries, unilateral and minimally displaced facet fractures without any sign of a spinal cord injury would be directed to non-operative management. However, the failure rate of non-operative treatment varies from 20 to 80%, and no consensus exists with regard to predictors of failure after non-operative management. CASE PRESENTATION: Case 1 is a patient with a unilateral facet fracture. The patient had only numbness in the right C6 dermatome but failed non-operative treatment, which resulted in severe spinal cord injury. Case 2 is a patient who had a similar injury pattern as case 1 but presented with immediate instability and underwent fusion surgery. Both patients had a minimally displaced unilateral facet fracture accompanied by disc injury and blunt vertebral artery injury, which are possible signs indicating significant instability. CONCLUSIONS: This is the first report of an isolated unilateral facet fracture that resulted in catastrophic spinal cord injury. These two cases illustrate that an isolated minimally displaced unilateral facet fracture with disc injury and vertebral artery injury were associated with significant instability that can lead to spinal cord injury.

Background and purpose - Deep-learning approaches based on convolutional neural networks (CNNs) are gaining interest in the medical imaging field. We evaluated the diagnostic performance of a CNN to discriminate femoral neck fractures, trochanteric fractures, and non-fracture using antero-posterior (AP) and lateral hip radiographs. Patients and methods - 1,703 plain hip AP radiographs and 1,220 plain hip lateral radiographs were included in the total dataset. 150 images each of the AP and lateral views were separated out and the remainder of the dataset was used for training. The CNN made the diagnosis based on: (1) AP radiographs alone, (2) lateral radiographs alone, or (3) both AP and lateral radiographs combined. The diagnostic performance of the CNN was measured by the accuracy, recall, precision, and F1 score. We further compared the CNN's performance with that of orthopedic surgeons. Results - The average accuracy, recall, precision, and F1 score of the CNN based on both anteroposterior and lateral radiographs were 0.98, 0.98, 0.98, and 0.98, respectively. The accuracy of the CNN was comparable to, or statistically significantly better than, that of the orthopedic surgeons regardless of radiographic view used. In the CNN model, the accuracy of the diagnosis based on both views was significantly better than the lateral view alone and tended to be better than the AP view alone. Interpretation - The CNN exhibited comparable or superior performance to that of orthopedic surgeons to discriminate femoral neck fractures, trochanteric fractures, and non-fracture using both AP and lateral hip radiographs.