Sparse Bayesian Learning for Label Efficiency in Cardiac Real-Time MRI

2025-03-27

AI Summary

In real-time cardiac MRI (≥50 fps), manual annotation of outer-slice frames is costly, and deep learning models suffer from poor generalization across slices. Method: This paper proposes a label-efficient modeling framework based on Sparse Bayesian Learning (SBL). We introduce SBL for the first time to model ventricular volume time-series signals, exploiting their spectral sparsity to accurately predict outer-slice volumes using annotations from inner slices only. We further design an active labeling strategy that minimizes posterior variance, with theoretical guarantees for its greedy optimization. The framework inherently provides uncertainty quantification. Results: Evaluated on real patient data, our method significantly reduces annotation burden, eliminates redundant annotations, and simultaneously improves both volumetric prediction accuracy and cross-slice robustness—demonstrating superior generalization with minimal supervision.

Technology Category

• Machine Learning: Active Learning
• Computer Vision: Multi-modal Vision
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Cardiac real-time magnetic resonance imaging (MRI) is an emerging technology that images the heart at up to 50 frames per second, offering insight into the respiratory effects on the heartbeat. However, this method significantly increases the number of images that must be segmented to derive critical health indicators. Although neural networks perform well on inner slices, predictions on outer slices are often unreliable. This work proposes sparse Bayesian learning (SBL) to predict the ventricular volume on outer slices with minimal manual labeling to address this challenge. The ventricular volume over time is assumed to be dominated by sparse frequencies corresponding to the heart and respiratory rates. Moreover, SBL identifies these sparse frequencies on well-segmented inner slices by optimizing hyperparameters via type -II likelihood, automatically pruning irrelevant components. The identified sparse frequencies guide the selection of outer slice images for labeling, minimizing posterior variance. This work provides performance guarantees for the greedy algorithm. Testing on patient data demonstrates that only a few labeled images are necessary for accurate volume prediction. The labeling procedure effectively avoids selecting inefficient images. Furthermore, the Bayesian approach provides uncertainty estimates, highlighting unreliable predictions (e.g., when choosing suboptimal labels).

Problem

• Improves label efficiency in cardiac real-time MRI segmentation
• Reduces manual labeling for outer slice ventricular volume prediction
• Ensures accurate volume prediction with minimal labeled images

Innovation

• Sparse Bayesian Learning for label efficiency
• Optimizes hyperparameters via type-II likelihood
• Provides uncertainty estimates for predictions