Artificial Intelligence in Renal Oncology: CNN-Based Classification of Kidney Cancer

Abstract

Medical researchers worldwide identify kidney cancer as a major mortal and morbid disease while its early-stage detection significantly improves patient survival potential. Global diagnostic practices based on imaging and biopsy techniques carry both costly intrusiveness and potential human errors in their diagnostic results. This research creates an automatic kidney cancer diagnostic system which utilizes deep learning CNN models and applies ResNet transfer learning techniques. Through the use of pre-trained ResNet architectures the model demonstrates robustness for classifying benign versus malignant kidney tumor images found in CT scans. The training process utilizes a kidney CT image dataset that required various preprocessing steps for both data augmentation and normalization and size adjustment to create a generalized model. The deep learning framework showcases superior performance than standard approaches while delivering a 92% accuracy rate through productive F1-scores alongside precise recall metrics. The Area Under the Curve metric of 0.95 demonstrates the model's powerful discriminatory performance. The utilization of ResNet transfer learning enabled models to extract meaningful insights from restricted datasets thereby minimizing dependence on lengthy labeled datasets while preserving prediction accuracy. The study demonstrates how CNN-based techniques present a promising diagnostic framework for clinical kidney cancer detection which combines accuracy with efficiency and non-invasiveness for aiding radiologists and clinicians during diagnostics.