Automatic classification of kidney stone components based on smartphone microscopy and the GoogLeNet model

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Urinary stone composition analysis is crucial for effective treatment.
• Current methods can be costly and time-consuming.
• An automated, accessible system is needed for rapid stone analysis.

Purpose of the Study:

• To develop an automated classification system for urinary stone composition.
• Integrate smartphone-based microscopic imaging (TIPSCOPE) with deep learning (GoogLeNet architecture).
• Enable rapid, accurate, and cost-effective analysis of stone composition.

Main Methods:

• Collected and classified 140 kidney stone samples into four types.
• Acquired microscopic images using TIPSCOPE and a smartphone.
• Trained a GoogLeNet classification model on 840 images (90% training, 10% testing).

Main Results:

• The system achieved an overall accuracy of 85.7% for stone composition classification.
• High F1 scores were observed for uric acid (0.92) and magnesium ammonium phosphate hexahydrate (0.95) stones.
• Calcium oxalate (0.86) and carbonate apatite (0.69) stones also showed promising classification results.

Conclusions:

• A novel system integrating smartphone microscopy and deep learning for kidney stone classification was successfully developed.
• The system demonstrates high accuracy and efficiency, particularly for specific stone types.
• This portable, low-cost solution presents a practical diagnostic tool for resource-limited healthcare settings.