Spatial Attention-guided Deep Learning for Accurate Kidney Disease Classification in CT Scans

Kidney diseases are globally prevalent and often lead individuals to seek urgent medical attention due to severe discomfort. The timely identification of these conditions necessitates a reliable automated diagnostic approach. Imaging techniques are integral in the diagnosis of kidney diseases, frequently requiring specialized expertise for comprehensive interpretation. Leveraging deep learning methodologies for kidney disease detection can offer valuable support to medical practitioners during diagnosis. This investigation delved into the categorization of kidney conditions based on CT scan images encompassing four distinct classes: cysts, normal, stone, and tumor. To facilitate this, a deep learning framework with spatial attention was introduced. Spatial attention modules were seamlessly integrated within Convolutional Neural Network (CNN) architecture to bolster interpretability, offering insights into influential regions guiding the model’s decisions and enhancing the model’s capacity to effectively process visual data, leading to improved classification performance. The proposed spatial attention-guided CNN model exhibited remarkable accuracy at 99.52%, accompanied by the precision, recall, and f1-score metrics of 99.54%, 99.52%, and 99.52% respectively. Impressively, the proposed framework surpassed the majority of recent comparable studies in terms of both accuracy and computational efficiency.