Novel Use of Deep Learning for Adaptive Weighted Loss in the Segmentation of X-ray Images for Detecting Wrist Fractures

Abstract

Image transmission technology has made significant contributions to various fields, especially medical imaging, which employs X-ray imaging to detect abnormalities in the human body, such as cracks and fractures. The image segmentation method is a computational method that can divide digital images into several segments with similar visual characteristics. This research proposes a novel image segmentation method for wrist fractures, utilising an adaptive weighted-loss approach. The objective is to enhance the precision of identifying wrist fractures on X-ray images, with the target being an accuracy of 71%, a precision of 78%, a recall of 85%, and an F1-score of 81%. This study addresses a significant gap in the literature by proposing an effective method for identifying wrist fractures, a task for which there has been a dearth of effective methods. The adaptive weighted-loss method with linear weights proposed in this study is expected to overcome the challenges posed by overfitting and correct class imbalance, thereby ensuring that the model is not overly dependent on a single loss function. This approach is anticipated to enhance the model's generalisation capabilities on test data.