Deep learning classification of pediatric spinal radiographs for use in large scale imaging registries.

PURPOSE: The purpose of this study is to develop and apply an algorithm that automatically classifies spine radiographs of pediatric scoliosis patients. METHODS: Anterior-posterior (AP) and lateral spine radiographs were extracted from the institutional picture archive for patients with scoliosis. Overall, there were 7777 AP images and 5621 lateral images. Radiographs were manually classified into ten categories: two preoperative and three postoperative categories each for AP and lateral images. The images were split into training, validation, and testing sets (70:15:15 proportional split). A deep learning classifier using the EfficientNet B6 architecture was trained on the spine training set. Hyperparameters and model architecture were tuned against the performance of the models in the validation set. RESULTS: The trained classifiers had an overall accuracy on the test set of 1.00 on 1166 AP images and 1.00 on 843 lateral images. Precision ranged from 0.98 to 1.00 in the AP images, and from 0.91 to 1.00 on the lateral images. Lower performance was observed on classes with fewer than 100 images in the dataset. Final performance metrics were calculated on the assigned test set, including accuracy, precision, recall, and F1 score (the harmonic mean of precision and recall). CONCLUSIONS: A deep learning convolutional neural network classifier was trained to a high degree of accuracy to distinguish between 10 categories pre- and postoperative spine radiographs of patients with scoliosis. Observed performance was higher in more prevalent categories. These models represent an important step in developing an automatic system for data ingestion into large, labeled imaging registries.

Does radiographic calibration affect linear radiographic measurements in a large pediatric spine registry?

PURPOSE: Large registries are increasingly at the forefront of modern pediatric spine research, with manual, centralized, trained radiographic measurement serving as the gold standard for spine research. However, there is limited data regarding the reliability of registry measurements which may be subject to differences in radiographic calibration. We undertook this study to evaluate reliability of T1-T12 height, L1-S1 height, and coronal balance measurements for a large registry of early onset scoliosis patients. METHODS: Three trained technicians from the Pediatric Spine Study Group measured 43 radiographs for T1-T12, L1-S1, and coronal balance using 3 different calibration techniques. All radiographs were AP views of patients with magnetically controlled growing rods with known diameters. The calibration techniques used a pre-export manually drawn line, a digital automatically generated computerized marker, and the diameter of a magnetically controlled growing rod. The intraclass correlation coefficient (ICC) was calculated to determine reliability. RESULTS: 1161 measurements were performed. For each of the three raters, coronal balance, T1-T12 height and L1-S1 height had excellent agreement regardless of the calibration technique (alpha 0.93-1.0). Among the parameters, coronal balance had the worst inter-rater reliability, whereas there was excellent interrater reliability regarding T1-T12 height and L1-S1 height (alpha 0.91-0.99). CONCLUSION: There was excellent agreement among reviewers and between the 3 different calibration techniques. While calibration using rod diameter served as the gold standard, this data shows that other standard calibration methods were adequate and achieved excellent reliability for registry radiographs.

Long-Term Outcomes at Skeletal Maturity of Combined Pelvic and Femoral Osteotomy for the Treatment of Legg-Calve-Perthes Disease.

Surgical treatment for Legg-Calve-Perthes disease (LCPD) is recommended for older children with moderate to severe disease. We sought to determine whether double osteotomies lead to improved radiologic outcomes compared to reported non-operative outcomes. Patients older than 6 years of age diagnosed with LCPD lateral pillar B or C who were treated with pelvic and femoral osteotomies were included. Radiologic outcomes and leg-length discrepancies were assessed using the Stulberg classification and were compared with the current literature. Fifteen hips in fourteen patients were treated with double osteotomy for LCPD, and seven had lateral pillar C disease (47%). The mean age at surgery was 8.6 years (range, 7.2-10.4) and the mean age at follow-up was 20.2 years (range, 14.2-35.6). At a mean 11.6-year follow-up (range: 6.3-25.2), double osteotomy resulted in 40% of patients having Stulberg I/II scores, 27% having Stulberg III scores, and 33% having Stulberg IV/V scores. The mean leg-length discrepancy was 1.4 cm in lateral pillar C patients compared to 0.8 cm in lateral pillar B patients. Four patients underwent additional surgeries, including two who required total hip arthroplasty. Double osteotomy as an alternative surgical procedure for the treatment of LCPD did not show improved outcomes when compared to historic non-operative cohorts.

Minimum 5-Year Results of Elongation Derotation Flexion Casting for Early Onset Scoliosis: The Story Is Not Over Until Skeletal Maturity.

BACKGROUND: Early conservative treatment for patients with idiopathic infantile scoliosis (IIS) with elongation derotation flexion (EDF) casting and subsequent serial bracing has become widely utilized. However, the long-term outcomes of patients treated with EDF casting are limited. METHODS: We performed a retrospective chart review of all patients who had undergone serial elongation derotation flexion casting and subsequent bracing for scoliosis presenting at a single large tertiary center. All patients were followed for a minimum of 5 years or until surgical intervention. RESULTS: Our study included 21 patients diagnosed with IIS and treated with EDF casting. At a mean 7-year follow-up, 13 of the 21 patients were considered successfully treated with a mean final major coronal curvature of 9 degrees compared to a pretreatment coronal curve of 36 degrees. These patients, on average, began casting at 1.3 years old and spent 1 year in a cast. Patients that did not have substantial improvement began casting at mean 4 years old and remained in a cast for 0.8 years. Three patients initially had substantial improvement with the correction to <20 degrees at a mean age of 7; however, their curves worsened in adolescence with poor brace compliance. All 3 patients will require surgical intervention. Of the patients not successfully treated with casting, 7 required surgery at a mean 8.2 years of age, 4.3 years after initiation of casting. A significant predictor of treatment failure was older age of cast initiation ( P <0.001). CONCLUSIONS: EDF casting can be an effective cure for IIS patients if initiated at a young age with 15 of 21 patients successfully treated (76%). However, 3 patients had a recurrence in adolescence resulting in an overall success rate of only 62%. Casting should be initiated early to maximize the likelihood of treatment success and periodic monitoring should be continued through skeletal maturity as recurrence during adolescence can occur.