AI-Powered Early Detection of Kidney Stones Using a Hybrid CNN-LSTM Model

Abstract

This study introduces a sophisticated kidney stone detection system that combines Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to enhance diagnostic accuracy and improve the speed of medical responses. The model was developed using a diverse dataset of 8,755 ultrasound images, which consists of 4,341 images showing kidney stones and 4,414 normal ones. By utilizing the CNN's advanced feature extraction capabilities, the system effectively identifies and classifies kidney stones based on their size and spatial relationships within the renal anatomy. The model achieved an accuracy of around 97%, supported by robust precision and recall rates, demonstrating its efficacy in clinical practice. The performance evaluation employed a confusion matrix, along with binary cross-entropy loss, to ensure model consistency and stability. In addition to its detection capabilities, this system provides insights into kidney health by calculating risk scores for stone development and forecasting the likelihood of recurrence, making it a practical tool for healthcare providers. Visual examples of ultrasound images further validate the model’s proficiency in distinguishing between healthy and diseased conditions, highlighting its potential to enhance diagnostic methods in urology.