Automatic detection of bone fractures in x-ray images using deep learning methods

In this thesis, a deep learning based approach is proposed for the automatic detection of bone fractures in X-ray images. Since the assessment of bone fractures by radiologists is time-consuming and prone to human error, automatic fracture detection systems offer significant potential for reducing clinical workload and improving diagnostic accuracy. To this end, a hybrid classification method based on a Swin Transformer enhanced with CNN models is introduced and evaluated on the publicly available FracAtlas dataset, which contains 717 fracture and 3,366 non-fracture X-ray images. The study first compares the baseline performance of state-of-the-art CNN-based models. Then, after conducting an extensive hyperparameter optimization, the three best-performing models are identified, and their extracted features from these models are integrated into the Swin Transformer architecture. To evaluate the performance of the proposed method and ensure valid comparison with other studies in the literature, experimental results were obtained on the official test set provided with the FracAtlas dataset. The proposed method achieved superior performance with an accuracy of 94.28% compared to studies that reported experimental results on the FracAtlas original test set. The obtained results demonstrate that the proposed hybrid classification approach exhibits superior performance in bone fracture classification compared to baseline CNN-based methods.