Detection of incomplete atypical femoral fracture on anteroposterior radiographs via explainable artificial intelligence

Abstract

One of the key aspects of the diagnosis and treatment of atypical femoral fractures is the early detection of incomplete fractures and the prevention of their progression to complete fractures. However, an incomplete atypical femoral fracture can be misdiagnosed as a normal lesion by both primary care physicians and orthopedic surgeons; expert consultation is needed for accurate diagnosis. To overcome this limitation, we developed a transfer learning-based ensemble model to detect and localize fractures. A total of 1050 radiographs, including 100 incomplete fractures, were preprocessed by applying a Sobel filter. Six models (EfficientNet B5, B6, B7, DenseNet 121, MobileNet V1, and V2) were selected for transfer learning. We then composed two ensemble models; the first was based on the three models having the highest accuracy, and the second was based on the five models having the highest accuracy. The area under the curve (AUC) of the case that used the three most accurate models was the highest at 0.998. This study demonstrates that an ensemble of transfer-learning-based models can accurately classify and detect fractures, even in an imbalanced dataset. This artificial intelligence (AI)-assisted diagnostic application could support decision-making and reduce the workload of clinicians with its high speed and accuracy.

Figure 1

Schematic diagram of fracture detection from the radiograph.

Figure 2

Schematic diagram of (a) transfer learning and (b) ensemble method. Transfer learning is performed for each of the six pre-trained models, and the three models (models a, b, and c) having the highest accuracy are selected for the ensemble.

Figure 3

Accuracy and loss graphs for (a) EfficientNet B5, (b) EfficientNet B6, (c) EfficientNet B7, (d) DenseNet 121, (e) MobileNet V1, and (f) MobileNet V2. The red lines indicate the result for the training set, and the blue lines indicate the result for the validation set.

Figure 4

Training result of the models ROC curve and its AUC for the ensemble method.

Figure 5

Five atypical femoral fracture predictions from the validation set. The green box indicates ground truth labeled by the radiologist. The results of three single classifiers (EfficientNet B5, B7, and MobileNet V1) and the ensemble results are represented.

Figure 6

Comparison of accuracy based on the application of the Sobel filter and CLAHE. When the Sobel filter was applied to preprocess the radiographs, the accuracy was improved for EfficientNet B5, EfficientNet B7, and MobileNet V1 by 4.8%, 3.1%, and 5.8%, respectively, while CLAHE showed accuracy improvement of 2.9%, 2.4%, and 3.4%, respectively.

Figure 7

Confusion matrix for (a) validation set and (b) test set.