Attention Based Deep Neural Network for Classification of Kidney Ailments Using CT Images
Rashi Chauhan, Mohan Karnati, Pradeep Kumar Singh
Abstract
Kidney disease, a significant health issue affecting millions worldwide, requires an accurate early diagnosis to improve patients’ health. Using the latest developments in deep learning (DL), this study attempts to create a robust CNNbased model for predicting and classifying kidney ailments using computed tomography (CT) images. The dataset we used included 12,446 images divided into four classes: cysts, tumors, stones, and normal samples. Despite convolutional neural networks (CNNs) having shown promise in diagnosis, we need to improve their accuracy, practical application, and efficiency. This study proposes a unique approach for diagnosing kidney problems by integrating a CNN model with a dual attention mechanism to extract the critical and essential traits from the CT images. The Adam optimizer trained the model and evaluated it on several parameters, including accuracy, loss, precision, and recall. Additionally, our CNN model with a dual attention mechanism attained an impressive $\mathbf{9 9. 4 8 \%}$ accuracy, exceeding other conventional methods. The proposed model has a mere 0.22 million parameters, in contrast to conventional CNN models that necessitate a huge number of parameters to comprehend the underlying patterns of CT images. This not only prolongs the training period but also poses challenges for deploying the model on compact devices.