Deep Learning-Driven Multiclass Bone Fracture Detection and Localization: A Comparative Study of CNN Architectures with YOLOv8-based Segmentation in Medical Imaging

Abstract

Bone fracture is a common health complication and usually needs accurate diagnosis and early intervention. X-ray imaging is a common technique in diagnosing and categorizing such fractures, whereas manual analysis may be time consuming and may not be accurate. This paper actually describes a multiclass classification system based on the Convolutional Neural Networks (CNNs) and YOLOv8 that can be used to segment the image with the goal of fractured bone detection; that is, to automate the bone fracture diagnosis. We tested CNN models of EfficientNetV2, Xception, and MobileNetV3 on an image dataset of X-rays and obtained results of 93.49, 88.14, and 89.42 with EfficientNetV2 which is the best among the models. Also, it used YOLOv8 to perform the segmentation task and effectively localized and identified the bone fractures in the images of X-Rays. The integration of these methods yielded great enhancements in the classification as well as the localization and offered an efficient and accurate means of detecting and segmenting bone fractures. This study has shown the possibility of using a combination of deep learning models to analyze bone fracture and thus help improve the process of diagnosis in clinical and medical settings.

Ähnliche Arbeiten