Long bone fracture type classification for limited number of CT data with deep learning
Keon Myung Lee, Sang‐Yeon Lee, Chan Sik Han, Seung Myung Choi · 2020
High-energy collisions and sports injuries may result in long bone fractures. To treat and manage fracture patients, it is necessary for trauma surgeons and physicians to first classify their fracture type because their treatment is different according to their type. Facture classification demands high level of expertise and careful examination. To assist the fracture treatment in medical practice, this paper proposes a convolutional neural network-based classification method to classify lower long bone fractures in which output class label can be multiple, data size is small compared to the number of class labels, and classes in the data set are imbalanced. It presents how to organize the deep network model, how to prepare the data, how to augment data, how to determine the classes from the network output, and how to evaluate the performance of model. It also explains the characteristics of the computed tomography (CT) image data. In the experiments, the proposed method showed 80.6% precision and 92.0% recall for a data set of CT fracture images.