Deep Learning-Based Analysis of X-Ray Images for Bone Fracture Detection Using Quantum Fourier Transform

In this work, an attempt is made to improve the delicate fractures identification in X-ray images with the help of a pre-trained ResNet50 included in the multitask schema with added use of quantum Fourier transform (QFT). The aim is to identify and locate fine fractures from millions of X-ray images at the case level to assist physicians in delivering early treatment. To compute the remote dependency relation of each position of continuous X-ray image sections and channels, we proposed a new deep learning based pre-trained method using quantum Fourier transform (QFT) as pre-processing. To address this issue, we integrated a pre-trained ResNet50, which has been previously reported to have a strong capability in feature extraction, with a multitask network and QFT, which were expected to improve the model’s capacity in detecting and identifying subtle fracture patterns. Here, the QFT was performed on the ResNet50 extracted feature maps to enhance the detection sensitivity and to demarcate the faint fractures. By combining our multitask network, ResNet50 for the first-step feature extraction and individual detection and segmentation branches, with the assistance of QFT. We utilized a dataset of 600 X-ray images, sourced from a publicly available database, divided into a training set (500 cases) and a test set (100 cases). In this work, the performance of the proposed models was measured against different pre-trained datasets, which shows our method is performing better than other methods. The significance of global awareness and frequency area when it comes to medical images. Therefore, the proposed scheme of the ResNet50-based multitask network developed with QFT is significantly enhanced for detecting fine fractures in X-ray images and paves the way for the practical application in clinical diagnosis. Future work will be aimed at the fine-tuning of the network’s general structure and the investigation of the impact of adding more clinical data for the purpose of improving detection accuracy.