DL based Technique for Detecting Renal Disorder using MobileNetV2

Proper diagnosis of kidney diseases including cysts, stones and tumors continues to be a key clinical issue. The paper will suggest a lightweight deep learning model to classify the images of CT scan images of kidney disease into four categories: Cyst, Stone, Tumor, and Normal. The suggested method aims at resource-aware and high-efficiency classification, based on the use of MobileNetV2 with ImageNet pre-trained weights, and fine-tuned with Adam optimizer. The imbalance in classes is solved by augmenting images, and each category has 1,377 samples. The final accuracy of the model is 95.95% and F1-score of 0.9596, being better than ResNet18, AlexNet and ShuffleNetV2 and having much lower computational cost. A web based Flask application is further extended to allow real-time clinical execution thus the proposed system is feasible in resource constrained medical settings.