HAMIL-QA: Hierarchical Approach to Multiple Instance Learning for Atrial LGE MRI Quality Assessment.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Research

Background:

• Accurate assessment of left atrial fibrosis using 3D Late Gadolinium Enhancement (LGE) Magnetic Resonance Imaging (MRI) is essential for managing atrial fibrillation.
• Current LGE MRI quality evaluation is challenged by patient movement, imaging variability, and the scarcity of expert annotations for automated deep learning models.

Purpose of the Study:

• To introduce HAMIL-QA, a novel hierarchical multiple instance learning (MIL) framework for automated quality assessment of LGE MRI scans.
• To address the limitations of existing deep learning models, including limited annotations and high computational costs, in LGE MRI quality evaluation.

Main Methods:

• Developed HAMIL-QA, a MIL framework with a hierarchical bag and sub-bag structure for targeted analysis and aggregated volume-level insights.
• Utilized a dataset of LGE MRI scans to train and evaluate the HAMIL-QA framework.

Main Results:

• HAMIL-QA demonstrated superior performance compared to existing MIL methods and traditional supervised approaches.
• Achieved higher accuracy, AUROC, and F1-Score in predicting LGE MRI scan quality.
• The hierarchical MIL approach effectively reduced annotation requirements and computational load.

Conclusions:

• HAMIL-QA offers a scalable and accurate solution for automating LGE MRI quality assessment.
• The framework's ability to focus on diagnostically critical features enhances the reliability of quality predictions.
• This advancement holds significant potential for improving diagnostic accuracy and patient outcomes in atrial fibrillation management.