A Hybrid Deep Learning Architecture for Prediction of Renal Diseases using Residual, Squeeze-Expand and Depth-Wise Convolutional Blocks

In predictive and preventive healthcare, machine learning and deep learning demonstrate a pivotal role in the early diagnosis of diseases. Kidneys are sophisticated biological filters and play a vital role in maintaining body homoeostasis. However, kidney failure leads to many critical and chronic diseases. Early diagnosis of these diseases is essential for better treatment options. The study proposes a hybrid deep learning architecture based on the ideas of residual, squeeze-expand, and depth-wise separable convolution blocks. A dataset with computer tomography scan images of patients suffering from various kidney ailments is used to train state-of-the-art deep learning models to classify these images. The proposed model exhibits superior performance over the models (VGG16, MobileNet, InceptionV3, ResNet50, and SqueezeNet) in terms of accuracy, precision, recall, F1-score, and AUROC. The hybrid architecture achieves a near-perfect average accuracy of 0.9992 and 0.9975 with five-fold and ten-fold stratified cross-validation respectively. The proposed model also shows a perfect AUROC score of 1.00 on an independent data of 800 images. Among the models evaluated, the proposed model shows the best performance, followed by InceptionV3 and MobileNet for the prediction of renal diseases. The proposed model aids clinicians in the early diagnosis of renal diseases.