A Three-Phase Explainable Deep Learning Approach for Reliable Wrist Fracture Identification from X-Ray Images

Wrist fractures present significant challenges in clinical diagnosis, often leading to treatment delays and compromised recoveries. Manual diagnosis is resource-intensive and prone to errors. To address these issues, we propose a unique DeepWristFNet and explainable AI-based model for end-to-end classification of bone fractures utilizing a small dataset of 193 wrist X-ray images to diagnose wrist fractures from X-ray images. The DeepWristFNet model is 20 layers deep, consisting of a fire module, a shufflenet unit, 18 convolutional, and two fully connected layers. We applied data pre-processing techniques such as data augmentation, image enhancement, and image resizing approaches to increase the number of images, enhance the quality of images, and resize images according to the image input size of the DeepWristFNet model. The proposed method comprised three phases. In the first phase, we performed training, validation, and testing in an end-to-end manner and achieved validation and testing accuracies of 99.04% and 87.93%, respectively. The testing is performed using unseen samples (10% of the data) from the dataset to verify the generalization ability of the proposed framework. In the second phase, we tried to improve the testing accuracy further. We extracted features using the proposed DeepWristFNet approach and applied the ReliefF feature selection algorithm to select the most essential features for the classification. Then, these features are given as input to ten classifiers. Out of ten classifiers, five classifiers, i.e., Efficient linear SVM, quadratic SVM, Narrow NN, wide NN, and medium NN, achieved 100% testing accuracy on unseen samples. In the third phase, we incorporate a pioneering Fuzzy Logic-Based Classification to precisely identify the "Foreground" and "Background" regions within the X-ray images, including the areas that may potentially correspond to fracture regions, enhancing the interpretability of the model. Further visualization is carried out utilizing a Grad-CAM that draws attention to the exact areas in the image that are fractured. Additionally, we evaluated how well our strategy performed against cutting-edge deep transfer learning models. The proposed DeepWristFNet model outperforms other existing methods, according to experimental findings on the dataset. This study emphasizes the promise of DeepWristFNet as an effective and precise method for classifying bone fractures in clinical settings, especially when using small datasets of medical image data.