Advanced Diagnosis of Nephrolithiasis, Renal Tumors, and Cysts Using Machine Learning and Deep Learning Techniques

Automated identification and categorization of renal pathologies, such as kidney cysts, nephrolithiasis and tumors, represents a significant step forward in the fields of medical imaging and diagnostic methods. An extensive dataset of 12,446 computed tomography (CT) images from various medical institutions in Dhaka, Bangladesh, including cyst (3,709), neoplasm (2,283), calculus (1,377), and normative (5,077) images is used in this investigation. The data was thoroughly assembled to retain data integrity and protect patient confidentiality by expunging all identifiable information. Data augmentation techniques were used on the dataset to mitigate class imbalance to 1,377 images per category in order to improve model's robustness and generalizability. We executed five distinct image classification algorithms: A hybrid CNN with Random Forest (RF), CNN, ResNet50, ResNet101, and VGG16 is used for the study. Each model underwent training utilizing an 80:20 splitting for training and testing. Normalization techniques are applied to promote effective convergence. We used ReLU activation for feature extraction and optimized the feature extraction with our CNN architecture, consisting of three convolutional layers followed by max pooling layers. The hybrid model combined CNN for high level feature extraction with a Random Forest classifier using the advantages that come with ensemble learning. Transfer learning was also used by using pre-trained ResNet and VGG models and fine tuning the terminal layers for specific classification goals. We find that the CNN and the hybrid CNN with RF achieved very high accuracy levels of 99.89%, while VGG16 achieved 99.0%. Accuracies of 93.93% and 93.56% were achieved with ResNet50 and ResNet101 respectively. The results of this research highlight the use of machine learning methodologies to improve diagnostic precision and the clinical workflow in medical imaging.