A Deep Learning-Based Decision Support System for Cholelithiasis in MRI Data.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Radiology
• Gallbladder Imaging Analysis

Background:

• Cholelithiasis (gallstones) poses significant health risks if not diagnosed promptly.
• Advancements in deep learning enable automated detection of gallstones in MRI.
• AI can reduce the time and resources needed for gallbladder evaluations.

Purpose of the Study:

• To develop an AI support system with a graphical user interface for detecting gallstones in gallbladder MRI.
• To reduce manual effort and time in identifying gallstones.
• To automatically locate and label gallbladders in T2-weighted axial MR images for gallstone detection.

Main Methods:

• Modified Mask Region Based Convolutional Neural Network (Mask R-CNN) for instance segmentation.
• Training and evaluation on 788 axial MR images, with radiologist-labeled segmentation.
• Focus on a single optimal slice for automated analysis to support radiologists.

Main Results:

• The modified Mask R-CNN model achieved up to 0.89 accuracy in gallstone detection.
• Squeeze and excitation (SE) modification improved classification accuracy.
• Image-level stone detection showed improved accuracy, precision, and specificity.

Conclusions:

• The modified Mask R-CNN model demonstrates significant potential for clinical application in gallstone detection.
• Automated analysis of gallbladder MRI using AI can support radiological diagnosis.
• The developed system offers a practical approach to improving diagnostic efficiency.