Optimizing Bone Fracture Diagnosis: A Hybrid Approach using EfficientNet-B3 and SVM for Accurate Classification

The identification of osseous fractures has significant importance for defining precise medical diagnosis and con- structing treatment plans, especially in emergent and orthopedic contexts. This work proposes a new classification framework for different types of hand fractures through the integration of the EfficientNet-B3 deep learning model with an SVM classifying kernel. The feature extractor uses the pre-trained and then fine-tuned EfficientNet-B3 model on hand X-ray images to obtain both higher level and fine-grained features required to accurately detect the fractures. Subsequent features are then again passed to an SVM classifier which defines an ideal decision boundary mapping which increases the model’s accuracy and interpretability. The dataset utilized in this study consists of 1,000 hand X-ray images obtained from different medical centers, which went through different preprocessing to enhance the image quality, as outlined below: Conversion of the images to grayscale benzylpenicilloyl hydrochloride—Gaussian blurring of the images, and edge detection on the images. In assessing the efficacy of the model, a variety of evaluative metrics, including accuracy, precision, recall, F1-score, and additional analytical tools, were employed, with the accuracy percentage reaching an impressive 93.5 percent in classification. A comparison of the proposed model using existing approaches is promising when it comes to the adaption of the data set into an automated system with a highly accurate classification of bone fracture, making it ideal for future clinical application.