Hybrid CNN-Attention-LSTM Architecture for Automated Kidney Stone Detection and Classification from Medical Imaging

There is an explicit abstract in the document. Cleaned up, it reads: There are many common diseases that affect millions of people worldwide, among them kidney stones (urinary tract stones). Their detection requires accurate and early diagnosis so that appropriate treatment can be carried out as quickly as possible, because delays may cause the patient’s condition to deteriorate. Traditional diagnostic methods also rely heavily on the radiologist, which is time-consuming. This paper presents a new deep learning hybrid architecture that combines convolutional neural networks (CNNs), attention mechanisms, and long short-term memory networks (LSTMs) to automatically detect and classify kidney stones based on medical images. The proposed model uses a CNN to extract spatial features, attention mechanisms to focus on disease-related regions, and an LSTM to capture sequential dependencies in multi-slice CT scans. The model was evaluated on a comprehensive dataset of 4,850 CT images and achieved 96.2% accuracy, 95.8% precision, 96.5% recall, and an F1 score of 96.1% in classification tasks. The attention mechanism improves interpretability by highlighting areas of interest, making the system clinically applicable. Comparative analysis with state-of-the-art methods including ResNet50, VGG16, and traditional CNN architectures shows superior performance, with an average AUC of 0.95 across four stone types (calcium oxalate, uric acid, struvite, and cystine), representing a performance gain of 3–5% over existing methods.