Hybrid EfficientNet B4 and SVM framework for rapid and accurate bone cancer diagnosis from X-rays.

The early and correct diagnosis of bone cancer is important for treating both primary and metastatic conditions effectively. Traditional imaging techniques, like CT, MRI, and X-ray scans, depend exclusively on manual review, which is time-consuming and prone to human errors. Recently, ML and DL have enabled automated diagnostic systems that are more accurate, reliable, and efficient. Still, many of the existing approaches using DL suffer from high computational complexity, overfitting, and limited availability of robust datasets. This work proposes a novel diagnostic model for bone cancer, called OsteoCancerNet, which combines EfficientNetB4 for feature extraction with a support vector machine using the RBF kernel for classification. EfficientNetB4 captures efficiently both quantitative and qualitative features from X-ray images, and the SVM ensures robust binary classification. Extensive experiments using a large dataset with 29,952 X-ray images demonstrate that OsteoCancerNet provides 98% precision, 97.47% recall, 98% accuracy, and a 98% F1-score, thus outperforming traditional machine learning, deep learning, and transfer learning methods. Of note, the model maintains fast inference times of 41 milliseconds per image, making it suitable for real-time clinical applications. By combining deep learning feature extraction with traditional machine learning classification, OsteoCancerNet provides an efficient, accurate, and practical approach for the early detection of bone cancer. This approach has the potential to aid radiologists in timely diagnosis, decrease workload, and improve treatment outcomes, thus underlining the advantages of integrating DL and ML techniques within medical imaging. Keywords: OsteoCancerNet; computer-assisted diagnosis; bone cancer diagnosis; EfficientNet B4 model; SVM model; X-ray image analysis.