COMPARISON OF CNN AND SVM FOR BONE CANCER DISEASE CLASSIFICATION BASED ON IMAGE DATA
TL;DR
Results confirm that the synergy between CNN hierarchical feature extraction and SVM classification robustness can significantly minimize the risk of false negatives, making it highly recommended as a reliable Computer-Aided Diagnosis (CAD) instrument to assist medical practitioners in early detection of bone cancer.
Abstract
This research is motivated by the high urgency of early diagnosis in bone cancer cases to reduce patient mortality rates. This study comparatively analyzes the performance of the Convolutional Neural Network (CNN) algorithm with ResNet50 architecture, Support Vector Machine (SVM), and their integration in a Hybrid CNN-SVM model for medical image classification. The research methodology involved a dataset of 8,814 radiological images processed through normalization and augmentation stages. In single-model testing, the end-to-end ResNet50 architecture achieved an accuracy of 87%, but showed limitations in generalizing microscopic textures at the softmax classification layer. On the other hand, the SVM algorithm supported by manual Histogram of Oriented Gradients (HOG) feature extraction demonstrated significant stability with an accuracy of 93.58%, proving the superiority of the optimal margin method in handling specific feature dimensions in medical images. The crucial finding in this study shows that the Hybrid CNN-SVM model—which utilizes ResNet50 as an automatic feature extractor and SVM as the final classifier—achieved peak performance with an accuracy of 95.18%, Precision value of 0.98, Recall of 0.96, and AUC of 0.98. These results confirm that the synergy between CNN hierarchical feature extraction and SVM classification robustness can significantly minimize the risk of false negatives, making it highly recommended as a reliable Computer-Aided Diagnosis (CAD) instrument to assist medical practitioners in early detection of bone cancer.