Quantitative image quality metrics enable resource-efficient quality control of clinically applied AI-based...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Quality Control

Background:

• AI-based MRI reconstruction accelerates imaging while preserving or enhancing image quality.
• Quality assessment of AI tools is recommended, but long-term monitoring faces challenges like model drift.
• Radiologist evaluations are resource-intensive and subjective, necessitating efficient quality control (QC) measures.

Purpose of the Study:

• To explore the use of image quality metrics (IQMs) for assessing AI-based MRI reconstructions.
• To evaluate the efficiency and effectiveness of IQMs as QC measures for AI MRI.

Main Methods:

• 58 patients underwent rectal MRI using AI-based and conventional T2-weighted sequences.
• Paired and unpaired IQMs were calculated and assessed for sensitivity to perturbations using control charts.
• Radiologists evaluated perturbed images to determine clinical relevance.

Main Results:

• Paired IQMs effectively detected deviations in AI reconstructions, outside ±2 standard deviations of the reference dataset.
• Unpaired metrics showed lower sensitivity compared to paired metrics.
• Paired IQMs showed no significant performance difference across 1.5 T and 3 T systems.

Conclusions:

• IQMs serve as effective and resource-efficient QC tools for AI-based MR reconstructions.
• IQMs offer a viable alternative to repeated, resource-intensive radiologist evaluations.
• Future research should extend IQM application to other imaging modalities and explore additional metrics.