Effective deep learning classification for kidney stone using axial computed tomography (CT) images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nephrology

Background:

• Kidney stones are a prevalent condition with high recurrence and morbidity rates.
• Computed tomography (CT) imaging is the preferred diagnostic method for kidney stones.
• Manual analysis of CT scans for kidney stone diagnosis is labor-intensive and time-consuming.

Purpose of the Study:

• To accurately classify kidney stones from CT scans using deep learning (DL) algorithms.
• To develop an automated system for kidney stone detection, reducing radiologist workload.

Main Methods:

• The Inception-V3 model was utilized as a reference architecture.
• Deep learning models were pre-trained using Convolutional Neural Network (CNN) architectures.
• The models were applied to a dataset of abdominal CT scans from patients with kidney stones.

Main Results:

• Eight DL models were evaluated on 8,209 CT images.
• The Inception-V3 model demonstrated a high test accuracy of 98.52% in detecting kidney stones.
• The study utilized a limited set of authentic CT images for training and testing.

Conclusions:

• The Inception-V3 model is effective in detecting small kidney stones.
• The model's high performance indicates its suitability for clinical applications.
• This DL approach can assist radiologists, reducing computational costs and the need for extensive expert review.