Deep learning computer vision algorithm for detecting kidney stone composition.

Area of Science:

• Nephrology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate identification of kidney stone composition is crucial for effective treatment.
• Current methods for stone analysis can be invasive or time-consuming.
• Digital photography offers a non-invasive approach to stone assessment.

Purpose of the Study:

• To evaluate the recall of a deep learning (DL) method for automatically detecting kidney stone composition from digital photographs.
• To assess the performance of a convolutional neural network (CNN) in classifying various kidney stone types.

Main Methods:

• Sixty-three human kidney stones of diverse compositions were analyzed.
• Digital images of stone surfaces and cores were captured.
• A ResNet-101 CNN model was employed for multi-class classification using leave-one-out cross-validation.

Main Results:

• The DL method achieved high recall rates for different stone compositions: uric acid (94%), calcium oxalate monohydrate (90%), struvite (86%), cystine (75%), and brushite (71%).
• The overall weighted recall for the CNN's composition analysis was 85%.
• High specificity and precision were observed across various stone types.

Conclusions:

• Deep convolutional neural networks demonstrate significant potential for identifying kidney stone composition from digital images with good recall.
• Further research is warranted to explore DL's application in detecting stone composition during digital endoscopy.
• This technology could lead to integrated endoscopic and laser systems for enhanced surgical efficiency.