A Hybrid Multi-Modal Deep Learning Framework for Automated Fracture Detection in Radiographs and CT Images

Abstract

Bone fractures are among the most prevalent musculoskeletal injuries, necessitating prompt and accurate diagnosis to ensure effective treatment and reduce complications.Traditional fracture detection relies heavily on the manual interpretation of X-ray and computed tomography (CT) images by radiologists, which is time-intensive and susceptible to human error, especially in the case of subtle or complex fractures.Recent studies have emphasized the crucial need for automated systems that can support clinicians in achieving faster and more accurate diagnoses without compromising patient safety.To address these challenges, this paper proposes FracturaX, a novel hybrid multi-modal deep learning framework designed for automated fracture detection across both X-ray and CT modalities.Unlike conventional models that rely on single imaging modalities, FracturaX capitalizes on the strengths of both 2D and 3D image data to enhance diagnostic depth.The proposed architecture integrates handcrafted radiomics features with deep convolutional features through a multistream network and an attention-based feature fusion mechanism, enhancing detection accuracy and robustness.This multistream approach not only improves fracture localization but also helps the model generalize across varied clinical settings.The framework was evaluated on diverse datasets, demonstrating superior performance compared to existing single-modality approaches and providing interpretable visual explanations to support clinical decision-making.Experimental results confirm that FracturaX offers a promising step toward reliable, generalizable, and explainable computer-aided fracture diagnosis, potentially reducing diagnostic workload and improving patient outcomes.

Ähnliche Arbeiten