Assessing the speed-accuracy trade-offs of popular convolutional neural networks for single-crop rib fracture classification.

Rib fractures are injuries commonly assessed in trauma wards. Deep learning has demonstrated state-of-the-art accuracy for a variety of tasks, including image classification. This paper assesses the speed-accuracy trade-offs and general suitability of four popular convolutional neural networks to classify rib fractures from axial computed tomography imagery. We transfer learned InceptionV3, ResNet50, MobileNetV2, and VGG16 models, additionally training "decomposed" models comprised of taking only the first n blocks for each block for each architecture. Given that acute (new) fractures are generally most important to detect, we trained two types of models: a classful model with classes acute, old (healed), and normal (non-fractured); and a binary model with acute vs. the other classes. We found that the first 7 blocks of InceptionV3 achieved the best results and general speed-accuracy trade-off. The classful model achieved a 5-fold cross-validation average accuracy and macro recall of 96.00% and 94.0%, respectively. The binary model achieved a 5-fold cross-validation average accuracy, macro recall, and area under receiver operator characteristic curve of 97.76%, 94.6%, and 94.7%, respectively. On a Windows 10 PC with 32GB RAM and an Nvidia 1080ti GPU, the model's average CPU and GPU per-crop inference times were 13.6 and 12.2 ms, respectively. Compared to the InceptionV3 Block 7 classful model, a radiologist with 9 years of experience was less accurate but more sensitive to acute fractures; meanwhile, the deep learning model had fewer false positive diagnoses and better sensitivity to old fractures and normal ribs. The Cohen's Kappa between the two was 0.813.

Early detection of ankylosing spondylitis using texture features and statistical machine learning, and deep learning, with some patient age analysis.

Ankylosing spondylitis (AS) is an arthritis with symptoms visible in medical imagery. This paper proposes, to the authors' best knowledge, the first use of statistical machine learning- and deep learning-based classifiers to detect erosion, an early AS symptom, via analysis of computed tomography (CT) imagery, giving some consideration to patient age in so doing. We used gray-level co-occurrence matrices and local binary patterns to generate input features to machine learning algorithms, specifically k-nearest neighbors (k-NN) and random forest. Deep learning solutions based on a modified InceptionV3 architecture were designed and tested, with one classifier produced by training with a cross-entropy loss function and another produced by additionally seeking to minimize validation loss. We found that the random forest classifiers outperform the k-NN classifiers and achieve an eightfold cross-validation average accuracy, recall, and area under receiver operator characteristic curve (ROC AUC) of 96.0%, 92.9%, and 0.97, respectively, for erosion vs. young control patients, and 82.4%, 80.6%, and 0.91, respectively, for erosion vs. old control patients. We found that the deep learning classifier trained without minimizing validation loss was best and achieves an eightfold cross-validation accuracy, recall, and ROC AUC of 99.0%, 97.5%, and 0.97, respectively, for erosion vs. all (combined young and old) control patients; this classifier outperforms a musculoskeletal radiologist with 9 years of experience in raw sensitivity and specificity by 8.4% and 9.5%, respectively. Despite the relatively small dataset on which we trained and cross-validated, our results indicate the potential of machine and deep learning to aid AS diagnosis, and further research using larger datasets should be conducted.