Automated classification of urinary stones based on microcomputed tomography images using convolutional neural...

Area of Science:

• Medical imaging analysis
• Artificial intelligence in medicine
• Urology

Background:

• Accurate classification of urinary stones is crucial for effective treatment planning.
• Current classification methods can be time-consuming and may require specialized expertise.
• Distinguishing between calcium, uric acid, and mixture stones is essential.

Purpose of the Study:

• To develop and validate an automated approach for classifying urinary stones.
• To utilize micro-CT imaging combined with a convolutional neural network (CNN) for stone classification.
• To differentiate between calcium, uric acid, and mixture types of urinary stones.

Main Methods:

• Micro-CT scanning of 30 urinary stones to generate 2,430 image slices.
• Classification of stone types using energy dispersive X-ray (EDX) spectra via scanning electron microscopy (SEM).
• Development and optimization of a 15-layer CNN model using training, validation, and test datasets.

Main Results:

• The CNN model achieved a validation accuracy of 0.9852.
• The optimized CNN model demonstrated a high test accuracy of 0.9959.
• The automated classification system exhibited a low classification error of 1.2%.

Conclusions:

• The proposed automated CNN-based approach effectively classifies urinary stones into calcium, uric acid, and mixture types.
• Micro-CT imaging coupled with CNNs offers a promising tool for urinary stone analysis.
• This automated method has the potential to improve the efficiency and accuracy of urinary stone diagnosis.