Kidney Ensemble-Net: Enhancing Renal Carcinoma Detection Through Probabilistic Feature Selection and Ensemble Learning

Kidney renal carcinoma is a type of cancer that originates in the renal cortex, the outer part of the kidney. It includes various subtypes, such as clear cell, papillary and chromophobe renal cell carcinomas, which are characterized by different cellular structures and behaviours. This cancer is often detected through imaging techniques and poses significant challenges due to its potential to metastasize and vary in treatment response. To address these challenges, we developed a novel computational framework named Kidney Ensemble-Net, designed to enhance the accuracy of renal carcinoma classification. Our approach begins by acquiring spatial features from contrast-enhanced images using a Convolutional Neural Network (CNN) effectively capturing intricate patterns and structures characteristic of different carcinoma subtypes. These extracted features are then transferred into a refined probabilistic feature set, upon which we construct an ensemble model leveraging the strengths of Logistic Regression (LR), Random Forest (RF), and Gaussian Naive Bayes (GNB) classifiers. The integration of these models within the Kidney Ensemble-Net architecture resulted in an outstanding performance, with our Kidney Ensemble-Net + LR model achieving a 99.72% accuracy score significantly surpassing existing state-of-the-art methodologies. Furthermore, we rigorously evaluated our model using k-fold validation analysis, ensuring its robustness and generalizability across diverse datasets. This comprehensive comparison with current leading approaches highlights the potential of Kidney Ensemble-Net as a powerful tool for the precise and reliable classification of kidney renal carcinoma, paving the way for improved diagnostic and treatment strategies.