Abstract
Kidney diseases have been identified as potentially fatal to human beings and this has prompted practitioners to recommend early diagnosis of the conditions. This study discusses an extensive overview on the classification of the kidney diseases using novel deep learning approaches such as CNN’s and Xception, Inception v3, EfficientNet-B2 models. Research exploits the collection of 12,446 of the CT images divided into classes based on the various kidney conditions including normality, cysts, tumor, and stones. A careful heuristics procedure is used in data pre-processing, which involved assigning labels, reshaping images and converting categorical into binary form. They are compared considering corresponding parameters such as precision, recall, F1-score and the data shows that EfficientNet-B2 model offers highest accuracy of 99 percent. Building a seamless web application with Flask where a user can easily upload images of the kidney CT, they will be classified without a problem. The results reveal that current state-of-the-art deep learning models can improve diagnostic performance and carry great practicable value in the field of nephrology. Hence, the future work is proposed to include a bigger sample size for the study, and moreover, apply more various types of imaging, and the improved methods of artificial intelligence based on explanation. This study aims to provide a new valuable contribution to the area of medical imaging and specifically bring in the use of machine learning into diagnosing and managing kidney diseases.
Keywords: Kidney Disease, Deep Learning, Xception, Inception v3, EfficientNet-B2