Abstract:

Early detection of kidney disease is essential for preventing severe complications and improving patient outcomes. Traditional diagnostic approaches, such as blood and urine tests, are invasive and time-consuming. This study explores a non-invasive iris-based diagnostic approach utilizing deep learning models to classify individuals as normal or abnormal based on kidney disease markers present in the iris. We implemented VGG16, ResNet50, Squeeze-DDConvNet, Separable CNN, and Depthwise CNN for feature extraction and classification. The models were trained on a dataset comprising iris images from healthy individuals and patients diagnosed with kidney disease. Image preprocessing techniques, including normalization and contrast enhancement, were applied to improve feature visibility. Among the architectures, Separable CNN outperformed others, achieving the highest classification accuracy for predicting abnormal and normal conditions. The model demonstrated superior feature extraction efficiency and reduced computational complexity compared to standard convolutional models.The findings suggest that Separable CNN-based iris analysis can serve as an effective, non-invasive screening tool for kidney disease detection. This approach holds promise for early diagnosis, reducing reliance on invasive tests and enabling widespread deployment in clinical and remote healthcare settings.

Introduction:

Kidney disease is a critical global health concern, affecting millions of individuals and contributing to high morbidity and mortality rates. It encompasses a range of disorders, from chronic kidney disease (CKD) to acute renal failure, often leading to life-threatening complications if left undiagnosed or untreated. Early detection and continuous monitoring are crucial to prevent progression to end-stage renal disease (ESRD), which necessitates dialysis or kidney transplantation. However, traditional diagnostic methods such as serum creatinine measurement, glomerular filtration rate (GFR) estimation, and urine protein analysis are invasive, time-consuming, and require well-equipped laboratory settings. These limitations highlight the need for non-invasive, rapid, and cost-effective diagnostic alternatives that can be deployed in both clinical and remote healthcare environments.
In recent years, biometric health assessment has gained significant attention, with the human iris emerging as a potential diagnostic marker for systemic diseases. The iris is a highly structured and unique biological feature containing complex vascular patterns, pigmentation, and crypts, which have been linked to underlying physiological and pathological changes in the body. Studies have demonstrated that alterations in iris texture, coloration, and vascular distribution may correlate with diseases such as diabetes, hypertension, and kidney dysfunction. Leveraging these ocular biomarkers through deep learning-based image analysis can offer a novel, non-invasive approach to kidney disease diagnosis.
Deep learning, particularly Convolutional Neural Networks (CNNs), has revolutionized medical image analysis by enabling automated feature extraction and classification with high precision. In this study, we evaluate the efficacy of multiple CNN architectures, including VGG16, ResNet50, Squeeze-DDConvNet, Separable CNN, and Depthwise CNN, for kidney disease detection based on iris images. These architectures are trained to distinguish between normal and abnormal iris patterns associated with renal dysfunction. Among these, Separable CNN achieved the highest classification accuracy, demonstrating its superior feature extraction capability, reduced computational complexity, and enhanced predictive performance compared to standard convolutional models.
The proposed system aims to provide a non-invasive, AI-driven diagnostic framework that can be integrated into clinical decision support systems, telemedicine platforms, and remote healthcare services. By automating the detection of kidney disease through iris recognition, this research contributes to early screening, timely intervention, and improved patient outcomes. Furthermore, the deployment of deep learning-based iris analysis can bridge the gap between advanced healthcare and under-resourced regions, making diagnostic solutions more accessible and efficient.