Enhanced Bone Fracture Detection through Deep Learning-Based Multi-Scale Feature Fusion

Abstract

Bone fractures, caused by external trauma or internal factors such as osteoporosis or bone cancer, pose significant challenges for accurate diagnosis and treatment planning. Traditional diagnostic methods, reliant on radiological imaging like CT scans, MRIs, and X-rays depend heavily on expert interpretation, which can be inconsistent and error-prone under high workloads. The advent of artificial intelligence (AI), particularly deep learning techniques, offers transformative solutions for automated, reliable, and efficient fracture detection and classification. This study introduces a deep learning model incorporating multi-scale feature fusion to elevate the detection and classification of bone fractures in radiological images. By leveraging parallel convolutional pathways with varying filter sizes, the model captures both fine-grained details, such as fracture lines, and broader contextual features like bone alignment and geometry. Comprehensive preprocessing techniques were applied to improve image quality and mitigate data limitations. Techniques like rotations, flips, zooms, and translations are utilized in Data augmentation techniques to synthetically enlarge and vary the training dataset. The model was evaluated on a diverse dataset, achieving an accuracy of 99.8%. The results demonstrate the model's reliability, precision, capacity for generalizing to unfamiliar data, offering a valuable tool to support radiologists in clinical settings. The integration of multi-scale feature fusion with CNN architectures not only addresses the complexities of fracture detection but also provides a scalable solution for diverse medical imaging challenges.

Ähnliche Arbeiten