Comparative Analysis of EfficientNet-V2 and InceptionNet-V2 for Accurate Kidney Disease Detection Using CT Scans

A comparative analysis has been performed on two deep learning models for the classification of kidney disease, namely EfficientNet-V2 and InceptionNet-V2. The performance evaluation is done with respect to four types of conditions: Tumor, Cyst, Normal, and Stone. In this study, a dataset of 2,490 images was used. The classification performance of each model was measured by their precision, recall, and F1-score. EfficientNet-V2 yielded an overall accuracy of 89% with a macro-average F1-score of 0.86, while performing exclusively better in detecting cysts: precision of 0.90, recall of 0.95, and F1-score of 0.93. Tumor detection is strong, with an F1-score of 0.91. But the model is relatively less effectual in classifying a kidney as normal, for which it had given an F1-score of 0.77. InceptionNet-V2 achieved 85% accuracy with an average macro F1-score of 0.81. It performed comparatively poorly in classifying normal conditions, reaching a low recall of 0.54 and an F1-score of 0.63. The case of cyst detection had slightly lower results than those from EfficientNet-V2, which yielded an F1-score of 0.89. These results show that both models behave well for the problem, but EfficientNet-V2 gave higher accuracy and good overall performance, especially in tumor and cyst case detection. These results indicate the hope of advanced deep learning techniques improving the current medical diagnostic process of kidney disease.