Enhancing Kidney Disease Classification through Transfer Learning with VGG16

This work introduces a thorough methodology for classifying kidney disease with deep learning techniques, specifically highlighting the well-recognized VGG16 model. The dataset used in this study was obtained from Kaggle and consists of a wide range of medical images depicting normal instances of kidney cysts, tumours, and stones. Each image has been standardised to a resolution of 150x150 pixels. The dataset is systematically partitioned into training and validation sets and then subjected to thorough preprocessing, normalisation, and augmentation techniques in order to optimise the training of the model. The utilisation of the VGG16 model architecture is implemented, incorporating base and intermediate convolutional layers to facilitate the recognition and extraction of fundamental features, complemented by fully connected layers to enable decisive classification. In order to optimise the VGG16 model for the purpose of classifying kidney diseases, a customised classification layer is incorporated, hence improving its effectiveness in the field of medical image analysis. This work aims to enhance the VGG16 model's accuracy to 98% through the adjustment of pre-existing weights, the establishment of non-trainable weights for the purpose of universal feature recognition, and the customisation of the model to effectively classify kidney conditions by accounting for their unique characteristics. The objective of this project is to create a highly accurate and dependable model for promptly detecting and categorising kidney problems. This research will make a substantial contribution to enhancing patient care in the fields of nephrology and urology.