Automated MR Image Prescription of the Liver Using Deep Learning: Development, Evaluation, and Prospective...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• There is a need for automated liver image prescription in MRI to enhance efficiency and reproducibility.
• Current manual prescription methods can be time-consuming and prone to variability.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) system for automated liver image prescription.
• To assess the accuracy and reproducibility of AI-driven liver MRI prescription compared to manual methods.

Main Methods:

• A convolutional neural network (YOLOv3) was trained on 1039 liver MRI localizer acquisitions with radiologist annotations.
• The AI model detected anatomical structures (liver, torso, arms) to generate 2D bounding boxes for automated 3D liver prescription.
• Performance was evaluated using Intersection over Union (IoU) for 2D detection and boundary mismatch for 3D accuracy, with prospective validation on healthy volunteers.

Main Results:

• The AI system achieved excellent agreement with manual annotations (median IoU > 0.91).
• Automated 3D prescription showed high accuracy, with shifts <2.3 cm in 99.5% of test datasets, comparable to interreader reproducibility.
• The AI performed well across various MRI sequences and field strengths (1.5T, 3.0T).

Conclusions:

• AI-based automated liver image prescription shows promising results for accuracy and reproducibility.
• This technology has the potential to significantly improve the efficiency of liver MRI examinations.
• The AI system demonstrated robust performance across diverse patient populations and imaging parameters.