Faculty Bibliography

Musculoskeletal trauma accounts for a significant fraction of emergency department visits and patients seeking urgent care, with a high financial cost to society. Diagnostic imaging is indispensable in the workup and management of trauma patients. However, diagnostic imaging represents a complex multifaceted system, with many aspects of its workflow prone to inefficiencies or human error. Recent technological innovations in artificial intelligence and machine learning have shown promise to revolutionize our systems for providing medical care to patients. This review will provide a general overview of the current state of artificial intelligence and machine learning applications in different aspects of trauma imaging and provide a vision for how such applications could be leveraged to enhance our diagnostic imaging systems and optimize patient outcomes.

Purpose/UNASSIGNED:To train convolutional neural network (CNN) models to classify benign and malignant soft-tissue masses at US and to differentiate three commonly observed benign masses. Materials and Methods/UNASSIGNED:= 227) were used to train and evaluate a CNN model to distinguish malignant and benign lesions. Twenty percent of cases were withheld as a test dataset, and the remaining cases were used to train the model with a 75%-25% training-validation split and fourfold cross-validation. Performance of the model was compared with retrospective interpretation of the same dataset by two experienced musculoskeletal radiologists, blinded to clinical history. A second group of US images from 275 of the 419 patients containing the three common benign masses was used to train and evaluate a separate model to differentiate between the masses. The models were trained on the Keras machine learning platform (version 2.3.1), with a modified pretrained VGG16 network. Performance metrics of the model and of the radiologists were compared by using the McNemar test, and 95% CIs for performance metrics were estimated by using the Clopper-Pearson method (accuracy, recall, specificity, and precision) and the DeLong method (area under the receiver operating characteristic curve). Results/UNASSIGNED:The model trained to classify malignant and benign masses demonstrated an accuracy of 79% (95% CI: 68, 88) on the test data, with an area under the receiver operating characteristic curve of 0.91 (95% CI: 0.84, 0.98), matching the performance of two expert readers. Performance of the model distinguishing three benign masses was lower, with an accuracy of 71% (95% CI: 61, 80) on the test data. Conclusion/UNASSIGNED:The trained CNN was capable of differentiating between benign and malignant soft-tissue masses depicted on US images, with performance matching that of two experienced musculoskeletal radiologists.© RSNA, 2020.

Purpose: Ultrasound is increasingly utilized as the first-line diagnostic evaluation of superficial soft tissue masses. With growing health care costs, there is increasing pressure to develop cost-effective methods to triage patients with palpable masses. Deep convolutional neural networks (CNNs) have demonstrated the ability to classify images with good accuracy. We hypothesize that using a limited dataset, a CNN can be trained to classify benign versus malignant soft tissue masses seen on ultrasound.
Material(s) and Method(s): Ultrasound exams from 227 patients were selected with up to two pairs of gray scale and Doppler images extracted per patient. Pairs of gray scale and Doppler images were concatenated to create a single image for a total of 344 combined images. Images from 49 patients (96 images) were withheld for a pathology enriched test set (56 benign and 40 malignant). The remaining 248 images were used to train a CNN using an 80/20 training-validation split with five-fold crossvalidation. The model was trained on Keras using a pretrained VGG-16 architecture on a Nvidia GTX 1070 GPU. The withheld test set was used for a reader study which consists of two experienced musculoskeletal radiologists to assess the performance of the model.
Result(s): The CNN achieved an average accuracy of 0.87+/-0.07 on fivefold cross validation. The best performing model in the five folds was selected for comparison against two musculoskeletal radiologists on the pathology enriched test data set. The model achieved an accuracy 0.73 on the test data set and an AUC of 0.78 which was comparable to the performance of the two musculoskeletal radiologists (0.76 and 0.65 accuracy).
Conclusion(s): Using a relatively small data set, a CNN can be trained to differentiate between benign and malignant soft tissue masses seen on ultrasound with its performance approaching that of two experienced musculoskeletal radiologists

OBJECTIVE:To develop a deep convolutional neural network (CNN) to automatically segment an axial CT image of the pelvis for body composition measures. We hypothesized that a deep CNN approach would achieve high accuracy when compared to manual segmentations as the reference standard. MATERIALS AND METHODS/METHODS:We manually segmented 200 axial CT images at the supra-acetabular level in 200 subjects, labeling background, subcutaneous adipose tissue (SAT), muscle, inter-muscular adipose tissue (IMAT), bone, and miscellaneous intra-pelvic content. The dataset was randomly divided into training (180/200) and test (20/200) datasets. Data augmentation was utilized to enlarge the training dataset and all images underwent preprocessing with histogram equalization. Our model was trained for 50 epochs using the U-Net architecture with batch size of 8, learning rate of 0.0001, Adadelta optimizer and a dropout of 0.20. The Dice (F1) score was used to assess similarity between the manual segmentations and the CNN predicted segmentations. RESULTS:The CNN model with data augmentation of N = 3000 achieved accurate segmentation of body composition for all classes. The Dice scores were as follows: background (1.00), miscellaneous intra-pelvic content (0.98), SAT (0.97), muscle (0.95), IMAT (0.91), and bone (0.92). Mean time to automatically segment one CT image was 0.07 s (GPU) and 2.51 s (CPU). CONCLUSIONS:Our CNN-based model enables accurate automated segmentation of multiple tissues on pelvic CT images, with promising implications for body composition studies.

Body composition entails the measurement of muscle and fat mass in the body and has been shown to impact clinical outcomes in various aspects of human health. As a result, the need is growing for reliable and efficient noninvasive tools to measure body composition. Traditional methods of estimating body composition, anthropomorphic measurements, dual-energy X-ray absorptiometry, and bioelectrical impedance, are limited in their application. Cross-sectional imaging remains the reference standard for body composition analysis and is accomplished through segmentation of computed tomography and magnetic resonance imaging studies. However, manual segmentation of images by an expert reader is labor intensive and time consuming, limiting its implementation in large-scale studies and in routine clinical practice. In this review, novel methods to automate the process of body composition measurement are discussed including the application of artificial intelligence and deep learning to tissue segmentation.

PURPOSE/OBJECTIVE:The aim of this study was to assess differences in interreader variability among radiologists after the implementation of a computer-assisted reporting (CAR) tool for the interpretation of degenerative disc disease on lumbar spine MRI. METHODS:Thirty lumbar spine MRI examinations were selected from the radiology database. Five fellowship-trained musculoskeletal radiologists evaluated each L4-L5 disc in a blinded fashion and reported the findings using a traditional free dictation approach. One month later, they reinterpreted the same discs using a web browser-based CAR tool in the same blinded fashion. The degrees of central canal stenosis, neural foraminal stenosis, and facet joint osteoarthritis; presence or absence of lateral recess stenosis; types of disc bulge or herniation; and herniation location using both methods were recorded. Percentage disagreement among the radiologists for each variable was calculated and compared using the Wilcoxon signed rank test. RESULTS:There was a statistically significant decrease among the five radiologists in percentage disagreement for neural foraminal stenosis (46% versus 35%, P = .0146) and facet joint osteoarthritis (45% and 22%, P < .0001) for reports created by free dictation compared with those created using the CAR tool. There was no statistically significant difference in interreader variability for the assessment of central canal stenosis, lateral recess effacement, disc herniation, disc bulge, or herniation location. CONCLUSIONS:Implementation of a CAR tool for the interpretation of degenerative changes on lumbar spine MRI decreases interreader variability in the assessment of neural foraminal stenosis and facet joint osteoarthritis.

OBJECTIVE:The objective of this study is to determine how a limited protocol MR examination compares to a full conventional MR examination for the detection of non-degenerative pathology such as acute fracture, infection, and malignancy. MATERIALS AND METHODS/METHODS:A sample of 349 non-contrast MR exams was selected retrospectively containing a 3:1:1:1 distribution of negative/degenerative change only, acute fracture, infection, and malignancy. This resulted in an even distribution of pathology and non-pathology. A limited protocol MR exam was simulated by extracting T1-weighted sagittal and T2-weighted fat-saturated (or STIR) sagittal sequences from each exam and submitting them for blinded review by two experienced musculoskeletal radiologists. The exams were evaluated for the presence or absence of non-degenerative pathology. Interpretation of the limited exam was compared to the original report of the full examination. If either reader disagreed with the original report, the case was submitted for an unblinded adjudication process with the participation of a third musculoskeletal radiologist to establish a consensus diagnosis. RESULTS:There were five false negatives for a sensitivity of 96.9 % for the limited protocol MR exam. Infection in the psoas, paraspinal muscles, and sacroiliac joint, as well as acute fractures in transverse processes and sacrum were missed by one or more readers. No cases of malignancy were missed. Overall diagnostic accuracy was 96.0 % (335/349). CONCLUSIONS:MR imaging of the lumbar spine limited to sagittal T1-weighted and sagittal T2 fat-saturated (or STIR) sequences has high sensitivity for the detection of acute fracture, infection, or malignancy compared to a conventional MR examination.

AIM/OBJECTIVE:To determine the utility of barium studies for detecting abnormalities responsible for recurrent weight gain after gastric bypass surgery. METHODS:A computerized search identified 42 patients who had undergone barium studies for recurrent weight gain after gastric bypass and 42 controls. The images were reviewed to determine the frequency of staple-line breakdown and measure the length/width of the pouch and gastrojejunal anastomosis. A large pouch exceeded 6 cm in length or 5 cm in width and a wide anastomosis exceeded 2 cm. Records were reviewed for the amount of recurrent weight gain and subsequent weight loss after additional treatment. RESULTS:Staple-line breakdown was present in 6/42 patients (14%) with recurrent weight gain. When measurements were obtained, 13/35 patients (37%) with recurrent weight gain had a large pouch, three (9%) had a wide anastomosis, and four (11%) had both, whereas 22/42 controls (52%) had a large pouch, one (2%) had a wide anastomosis, and two (5%) had both. Ten patients (24%) with recurrent weight gain underwent staple-line repair (n = 3) or pouch/anastomosis revision (n = 7). These 10 patients had a mean weight loss of 38.1 lbs versus a mean loss of 8.6 lbs in 19 patients managed medically. CONCLUSION/CONCLUSIONS:Only 14% of patients with recurrent weight gain after gastric bypass had staple-line breakdown, whereas 57% had a large pouch, wide anastomosis, or both. Not all patients with abnormal anatomy had recurrent weight gain, but those who did were more likely to benefit from surgical intervention than from medical management.