Leveraging deep learning for comprehensive classification of renal diseases: A transfer learning approach

Agniva Banerjee

Keywords: CNN, Kidney, image classification, deep learning, transfer learning

Abstract: The nightmare of cancer as a leading cause of premature deaths worldwide is becoming real and turns out to be one of the major problems of humanity nowadays. Cancer diagnostics at the early stage is Critical to cancer recovery and survival. In this context, renal diseases, including kidney cysts, stones, and tumors, pose significant global health challenges, affecting approximately 12\% of the population and contributing to chronic kidney disease (CKD). Notably, renal cancer ranks as the tenth most prevalent cancer type, accounting for 2.7\% of all cancer cases. This work presents a deep learning (DL) framework utilizing transfer learning (TL) for the early detection of renal diseases and categorizing the conditions into four binary classifications: Cyst\_vs\_Normal, Cyst\_vs\_Stone, Cyst\_vs\_Tumor, and Stone\_vs\_Tumor, allowing for a more specific understanding of each stage. By analyzing CT scans and microscopic histopathology images, the framework employs convolutional neural networks (CNNs) with pre-trained models to facilitate automatic and precise classification of renal conditions. Specifically, two CNN models ResNet-50 and EfficientNetV2 are implemented, providing a comprehensive analysis of each stage of the DL architecture. Comparative evaluations of training outcomes across various datasets revealed that EfficientNetV2 performed marginally better than ResNet-50, achieving an impressive testing accuracy of up to 100\% for all cases. These results underscore the effectiveness of the DL-based system and highlight its potential for widespread clinical application in renal disease diagnosis.

Primary Area: transfer learning, meta learning, and lifelong learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6325