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

TL;DR

This work addresses the challenge of automating quality assessment for 3D LGE MRI scans used to quantify atrial fibrosis, where limited expert annotations and image variability hinder standard supervised approaches. It introduces HAMIL-QA, a hierarchical MIL framework that analyzes volume-level quality through a two-stage process: sub-bag (slice) attention-based MIL to identify diagnostic patches, and a bag-level MIL that aggregates distilled slice features to predict volume quality. The method demonstrates superior accuracy, AUROC, and F1 compared with fully supervised 3D CNNs and existing MIL baselines on a 424-scan LGE-MRI QA dataset, while offering substantial computational efficiency by processing 2D patches rather than full volumes. The approach yields interpretable attention maps highlighting diagnostically relevant regions around the left atrium, supporting potential clinical deployment for standardized QA in fibrosis quantification and ablation planning.

Abstract

The accurate evaluation of left atrial fibrosis via high-quality 3D Late Gadolinium Enhancement (LGE) MRI is crucial for atrial fibrillation management but is hindered by factors like patient movement and imaging variability. The pursuit of automated LGE MRI quality assessment is critical for enhancing diagnostic accuracy, standardizing evaluations, and improving patient outcomes. The deep learning models aimed at automating this process face significant challenges due to the scarcity of expert annotations, high computational costs, and the need to capture subtle diagnostic details in highly variable images. This study introduces HAMIL-QA, a multiple instance learning (MIL) framework, designed to overcome these obstacles. HAMIL-QA employs a hierarchical bag and sub-bag structure that allows for targeted analysis within sub-bags and aggregates insights at the volume level. This hierarchical MIL approach reduces reliance on extensive annotations, lessens computational load, and ensures clinically relevant quality predictions by focusing on diagnostically critical image features. Our experiments show that HAMIL-QA surpasses existing MIL methods and traditional supervised approaches in accuracy, AUROC, and F1-Score on an LGE MRI scan dataset, demonstrating its potential as a scalable solution for LGE MRI quality assessment automation. The code is available at: $\href{https://github.com/arf111/HAMIL-QA}{\text{this https URL}}$

Figures (2)

• Figure 1: Overview of our proposed model. For illustrative purposes, we select M axial slices (with 4 depicted as an example) at random from an LGE MRI scan, treating each slice as an individual sub-bag. We then extract random cropped patches from each of these slices. These sub-bags are initially processed by the sub-bag module. Subsequently, the output from sub-bag module is used to generate feature vectors, which are then input into the bag module. It is important to note that the ground truth label for the bags remains consistent across both sub-bag module and bag module during the training phase.
• Figure 2: Heatmap of 2 scans by original LGE MRI image and by attention map of our model, respectively. In the second column, a highlighted red square marks the patch receiving the highest attention weight, with an enlarged view provided for clarity. Additionally, a red arrow on the original MRI images indicates the left atrium's position.