Enhancing diagnosis: ensemble deep-learning model for fracture detection using X-ray images.

Aim: Orthopedic trauma results in the injury of bone joints and tendons of the body. A radiologist reviews and monitors large numbers of radiographs daily, which can lead to the diagnostic error. Therefore, there is a need to automate the detection of bone fractures in X-ray images, particularly humerus bone fractures. In this paper, we have proposed an ensemble model that can detect the fracture in an x-ray image.

Materials And Methods: In this paper, we proposed an ensemble model designed for fracture detection in X-ray images. An ensemble model combines multiple diverse models to improve predictive accuracy and robustness by aggregating their individual predictions. The model leverages MobileNetV2, Vgg16, InceptionV3, and ResNet50, using histogram equalization for preprocessing and a Global Average Pooling layer for feature extraction. The entire humerus from the public Mura-v1.1 dataset is utilized for analysis, utilizing a single training-validation split. The dataset is divided into a ratio of 80:20 for experiments for the training and validation datasets.

Results: The proposed model outperformed the modified deep-learning models and achieved 92.96%, 91.62%, and 92.14% accuracy, recall, and F1 scores, respectively.

Conclusion: The ensemble model presented effectively automates bone fracture detection in X-ray images of the humerus, demonstrating superior performance compared to modified deep-learning models. A comparison has been made between a novel ensemble model and state-of-the-art models, bench-marking their performance. These findings underscore its potential for enhancing diagnostic accuracy and efficiency in orthopedic radiology.