An Investigative Framework for Bone Cancer Detection Using Deep Learning

Abstract

Bone cancer is a rare yet life-threatening disease that demands early and accurate diagnosis to improve treatment outcomes. Over the past few years, it has been demonstrated that machine learning and radiomics greatly assist in interpreting medical images. The study introduces a Hybrid Diagnostic Framework by merging radiomics analysis and deep learning for better bone cancer classification accuracy on MRI images. A Bone Cancer MRI Dataset made available on Kaggle was used, consisting of images that were either benign, malignant or normal in nature. Using PyRadiomics, features such as shape, intensity and texture were extracted and EfficientNet-B0 was used to learn deep semantic features. The final model, called a Hybrid-Ensemble, was created by fusing Support Vector Machine, Gradient Boosting Machine and Multi-Layer Perceptron. The results from many experiments indicated that this model performed much better than other models, delivering 97% accuracy, 96% precision, 97% recall, 96.5% F1-score and 0.98 AUC. The outcomes show that the hybrid method is reliable and has strong diagnostic capabilities for detecting bone cancer automatically. With this approach, radiologists find it easier to make decisions and it may be useful in regular clinical settings.