Lightweight Deep Learning Model for Kidney Stone Detection: A Custom VGG16 with Layer Freezing Approaches

The growing prevalence of kidney stone disease and the need for early, accurate diagnosis necessitate the development of automated detection models that can assist radiologists in improving diagnostic efficiency and accuracy. This paper contributes an innovative way for kidney stone detection utilizing a customized VGG16 model with frozen and unfrozen layers. By leveraging the deep learning capabilities of the VGG16 architecture, we aim to intensify the accuracy as well as efficiency of kidney stone detection from medical images. Through a comprehensive analysis, performance evaluation of the model, particularly focusing on the impact of freezing and unfreezing specific layers. The study utilizes a dataset comprising medical images of kidney stones, and we employ techniques such as transfer learning to adapt the pre-trained VGG16 model to this task. Our findings demonstrate the effectiveness of fine-tuning the model’s architecture by using the 6 layer frozen customized vgg16 model , which is lightweight and computationally faster. It requires less number of hyperparameters., leading to improved classification accuracy 99.96%. Additionally, we discuss the potential clinical implications of our approach, highlighting its significance in aiding healthcare professionals in timely and accurate diagnosis. Overall, this paper puts up to the growing body of literature on the application of deep learning in medical image analysis and underscores the importance of customized model architectures for specific diagnostic tasks.