CalcSeg: Confidence-aware 3D Latent Context Curriculum Learning for Myocardial Scar Segmentation in LGE-CMR

Forum topic · 小凯 · 2026-08-22

Summary

CalcSeg is a confidence-aware latent context curriculum learning framework for myocardial scar segmentation from single-stack late gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) imaging. Scar segmentation in LGE-CMR is clinically important but challenging due to low tissue contrast, diffuse and small scar regions, and limited 3D spatial context in single-stack acquisitions. The framework introduces a dynamic semi-supervised curriculum learning strategy that progressively expands training from easier to more challenging scar cases using a learned confidence-aware scoring function. This function combines errors of the predicted scar maps, quantified epistemic uncertainty, and scar burden estimates to automatically assess sample difficulty without manual labels. To compensate for the limited spatial context of single-stack acquisitions, CalcSeg employs a latent slice-wise self-attention module that captures inter-slice dependencies and infers 3D spatial representations from sparse 2D inputs. Evaluation on multi-center clinical LGE-CMR datasets shows that CalcSeg consistently outperforms competing methods. Paper: arXiv 2608.20305, by Nivetha Jayakumar, Hannah Kim, Amit R. Patel, and Miaomiao Zhang.

Paper Overview

Field: Computer Vision Authors: Nivetha Jayakumar, Hannah Kim, Amit R. Patel, Miaomiao Zhang Published: 2026-08-22 arXiv: 2608.20305

Abstract

Segmentation of myocardial scars from single-stack late gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) imaging is a long-standing and clinically important challenge, particularly in the presence of low tissue contrast, diffuse and small scar regions. These challenges are further compounded by limited 3D spatial context.

This paper proposes CalcSeg, a confidence-aware latent context curriculum learning framework that leverages fused 3D feature representations from single-stack 2D LGE-CMR images for robust scar segmentation.

Key Contributions

Dynamic semi-supervised curriculum learning: A strategy that progressively expands training from easier to more challenging scar cases using a learned confidence-aware scoring function.
Confidence-aware scoring function: Integrates errors of the predicted scar maps, quantified epistemic uncertainty, and scar burden estimates to automatically assess sample difficulty without requiring manual labels.
Latent slice-wise self-attention: Developed to compensate for the limited spatial context of single-stack acquisitions by capturing inter-slice dependencies and inferring 3D spatial representations from sparse 2D inputs.

Results

Evaluation on multi-center clinical LGE-CMR datasets demonstrates that CalcSeg consistently outperforms all competing methods.

--- *Automatically collected on 2026-08-22*