Towards Clinical Decision Support: CNN-SVM Models for Kidney Tumor Classification

Kidney cancer is listed among the leading causes of cancer deaths and that is why accurate staging of the disease and early diagnosis is so crucial. This research introduces a hybrid framework combining Convolutional Neural Networks (CNN) and Support Vector Machines (SVM) for the classification of kidney tumors into four categories: Some of these are Clear Cell Renal Cell Carcinoma (ccRCC), Papillary Renal Cell Carcinoma (pRCC), Chromophobe Renal Cell Carcinoma (chRCC), Benign Kidney Tumors. The CNN carried out feature extraction in an automatic manner, while the strength of the SVM was in classification using its impressive decision boundaries. Using the proposed model, the experiments were carried out on the 4000 labelled images of kidney tumor with the average accuracy of 93.75% which clearly outperforms standalone CNN with the accuracy of 91% as well as classic SVM with hand-crafted features equal to 85%. The proposed model provided precision equal to 94% in case of ccRCC, 91% for pRCC, 90% for chRCC, a well as 96% in case of Benign Tumor; while the recall of the model precisely unfolded as 92%, 89%, 88% and 94% respectively. The F1-scores ranged between 89 per cent and 95 per cent and suggested good consistency across all the parameters. Based on the confusion matrix results, there was little crossover between the two types of renal cancers, as well as between ccRCC and pRCC especially with 20 false positives and 18 false negatives. The ROC curve analysis also showed that the AUC for four classes being 0.94, 0.91, 0.90 and 0.96 respectively. The results further prove the performance of the CNN-SVM system of handling multi-Class imbalance and achieving high diagnostic performance. This paper demonstrates its applicability as a clinical decision support tool, aspiring to enhance the diagnostic results and effectiveness.