Deep Learning Pipeline for Fully Automated Myocardial Infarct Segmentation from Clinical Cardiac MR Scans

TL;DR

This study introduces a fully automated deep learning pipeline for infarct and MVO segmentation on clinical LGE CMR images, combining LV localization with a cascaded 2D‑3D network and an error-correcting mechanism to handle small, poorly contrasted scars. Quantitative analysis shows infarct segmentation closely matches manual measurements ( Dice ≈ 64% ; concordance in volume ≈ 0.90) and is vastly faster (3–5 seconds per patient on CPU) with no preprocessing. Qualitative expert evaluation reveals AI infarct segmentations are often preferred over manual contours, though MVO detection remains more accurate when performed by humans. Overall, the approach markedly increases efficiency while maintaining clinically acceptable accuracy, supporting potential clinical adoption for rapid infarct quantification, with room for improvement in MVO performance and cross-institution generalization.

Abstract

Purpose: To develop and evaluate a deep learning-based method that allows to perform myocardial infarct segmentation in a fully-automated way. Materials and Methods: For this retrospective study, a cascaded framework of two and three-dimensional convolutional neural networks (CNNs), specialized on identifying ischemic myocardial scars on late gadolinium enhancement (LGE) cardiac magnetic resonance (CMR) images, was trained on an in-house training dataset consisting of 144 examinations. On a separate test dataset from the same institution, including images from 152 examinations obtained between 2021 and 2023, a quantitative comparison between artificial intelligence (AI)-based segmentations and manual segmentations was performed. Further, qualitative assessment of segmentation accuracy was evaluated for both human and AI-generated contours by two CMR experts in a blinded experiment. Results: Excellent agreement could be found between manually and automatically calculated infarct volumes ($ρ_c$ = 0.9). The qualitative evaluation showed that compared to human-based measurements, the experts rated the AI-based segmentations to better represent the actual extent of infarction significantly (p < 0.001) more often (33.4% AI, 25.1% human, 41.5% equal). On the contrary, for segmentation of microvascular obstruction (MVO), manual measurements were still preferred (11.3% AI, 55.6% human, 33.1% equal). Conclusion: This fully-automated segmentation pipeline enables CMR infarct size to be calculated in a very short time and without requiring any pre-processing of the input images while matching the segmentation quality of trained human observers. In a blinded experiment, experts preferred automated infarct segmentations more often than manual segmentations, paving the way for a potential clinical application.

Figures (12)

• Figure 1: AI pipeline. Firstly, a smaller image stack is extracted out of the original data by segmentation of the left ventricle. Then an error correcting 2D-3D cascaded framework is used to perform multiclass segmentation on the left ventricle.
• Figure 2: Scatter plot (left) and Bland-Altman analysis (right) of infarct size as a percentage of the total myocardial volume determined automatically and manually. In the scatter plot, the dashed line indicates $100$ percent agreement, and the solid line represents the linear regression line.
• Figure 3: Examples of infarct segmentations, which include both optimal and faulty CNN and human-based segmentations. Expert ratings for the corresponding images are displayed on the right-hand side.
• Figure 4: Experts rating for manual and automatic LGE segmentations. All slices were classified into true negative, true positive, false negative, and false positive (left). For true positive segmentations, the raters had to provide more detailed feedback (middle) and finally compare the two segmentation methods (right).
• Figure 5: Experts rating for manual and automatic MVO segmentations. All slices were classified into true negative, true positive, false negative, and false positive (left). For true positive segmentations, the raters had to provide more detailed feedback (middle) and finally compare the two segmentation methods (right).