The MYOSAIQ Challenge: Myocardial Segmentation with Automated Infarct Quantification

Olivier Bernard; William A. Romero R.; Cyprien Bouton; Celia Goujat; Hang Jung Ling; Pierre-Marc Jodoin; Fumin Guo; Calder Sheagren

eess.IVAug 29, 2026

The MYOSAIQ Challenge: Myocardial Segmentation with Automated Infarct Quantification

Authors: Olivier Bernard, William A. Romero R., Cyprien Bouton, Celia Goujat, Hang Jung Ling, Pierre-Marc Jodoin, Fumin Guo, Calder Sheagren, +19 more

Organizations: INSA-Lyon, Universite Claude Bernard Lyon 1, UJM-Saint Etienne, CNRS, Inserm, CREATIS UMR 5220, U1294, F-42023, Saint Etienne, France · Institut Universitaire de France (IUF), France · Department of Computer Science, University of Sherbrooke, Sherbrooke, QC, Canada · Wuhan National Library of Optoelectronics, Huazhong University of Science and Technology, Wuhan, China · Sunnybrook Research Institute, University of Toronto, Toronto, Canada · National Heart and Lung Institute, Faculty of Medicine, Imperial College London, London, United Kingdom · Centre for Medical Image Computing, Department of Computer Science, University College London, London, United Kingdom · Fudan University, Shanghai, 200433, China · Gottfried Schatz Research Center: Medical Physics and Biophysics, Medical University of Graz, Graz, Austria · Institute of Computer Graphics and Vision, Graz University of Technology, Graz, Austria · Institute for Medical Informatics, Statistics and Documentation, Medical University of Graz, Graz, Austria · Instituto de Investigaci´on en Ingenier´ıa de Arag´on, Universidad de Zaragoza; Instituto de Investigaci´on Sanitaria de Arag´on (IIS Arag´on); CIBER-BBN, Zaragoza, Arag´on, Spain · Centre Hospitalier Annecy Genevois, France · Hˆopital Universitaire Cardiologique Louis Pradel, Bron, France · CarMeN INSERM U1060, Universite Claude Bernard Lyon 1, Lyon, France · Hˆopital Universitaire de Saint-Etienne, Saint-Etienne, France.

Abstract

Late gadolinium enhancement (LGE) cardiac magnetic resonance (MR) imaging is the modality of choice to assess myocardial infarction (MI) lesions. Nowadays MI volume quantification is not performed routinely in clinical practice. Numerous deep learning (DL) methods have been developed to automate the segmentation of the myocardium and infarct regions. However, most studies rely on relatively small datasets which typically undergo pre-processing steps to standardize images and focus on a specific phase of myocardial infarction following reperfusion therapy. These limitations have impeded the development of models that are generalizable across diverse conditions and thus suitable for routine clinical use. To advance research and establish benchmarks in generalizable learning for myocardial infarct quantification, this paper presents findings from the Myocardial Segmentation with Automated Infarct Quantification (MYOSAIQ) challenge. The dataset set up for the challenge combines 439 CMR volumes from two multicenter clinical trials, with representative data acquired in acute and chronic phases after acute MI. Data were acquired in 16 centers using MRI scanners from three different vendors. Six teams participated until the end of the challenge, employing various baseline models, data augmentation techniques, and confidence strategies. To enhance the significance of this study, we compare the challengers' results with those of fine-tuned foundation models. Our results indicate that well-designed UNet-based techniques outperform fully automatic foundation models for LGE MR segmentation. While the best methods achieve high-quality and stable delineations of the left ventricle and myocardium under various conditions, they remain improvable in accurately segmenting infarct regions.

Figures & tables

Figure 1: Reference segmentation and regions of interest: left ventricular cavity (red), myocardium (green), myocardial infarction (magenta) and microvascular obstruction (yellow).

		Territory	D8	M1	M12
Clinical metrics
Transmurality		LAD	$40\pm 16$	$27\pm 15$	$25\pm 16$
(Extent of a lesion across)	(%)	LCX	$29\pm 09$	$17\pm 09$	$12\pm 08$
the thickness of the muscle)		RCA	$24\pm 10$	$17\pm 09$	$12\pm 08$
		LAD	$34\pm 11$	$29\pm 13$	$24\pm 14$
Endocardial Surface	(%)	LCX	$33\pm 08$	$22\pm 12$	$19\pm 11$

Table 1: Descriptors of the myocardial infarction lesion categorized by territory and time post reperfusion therapy. The endocardial surface length and the lesion size are provided in proportion of the whole endocardial length and the myocardial area, respectively.

Figure 2: Infarct location of the LAD, LCX and RCA infarcts after 8 days, 1 month and 1 year post reperfusion (average across the subjects in each subgroup).

Figure 3: Distribution of the in-plane resolution and slice thickness for the training and test datasets across the three experimental configurations (D8, M1, and M12).

\Block 2-1 Method

\Block 1-3 LV

\Block 1-3 MYO

\Block 1-3 MI

\Block 1-5 MVO

Dice

ASSD (mm)

HD (mm)

Dice

ASSD (mm)

HD (mm)

Dice

ASSD (mm)

HD (mm)

Dice

ASSD (mm)

HD (mm)

0.928

\pm

0.031

0.5

\pm

0.2

7.2

\pm

2.4

0.840

\pm

0.036

0.4

\pm

0.2

9.0

\pm

2.8

0.669

\pm

0.177

1.0

\pm

1.7

19.8

\pm

14.4

0.537

\pm

0.190

1.2

\pm

1.0

19.5

\pm

14.6

10/10

15/15

\arrayrulecolor gray!50 \arrayrulecolor black P2

0.933

\pm

0.029

0.4

\pm

0.2

6.8

\pm

2.1

0.840

\pm

0.038

0.4

\pm

0.1

9.0

\pm

3.0

0.700

\pm

0.145

1.0

\pm

1.7

16.8

\pm

13.2

0.681

\pm

0.119

0.7

\pm

0.4

14.4

\pm

4.7

9/10

15/15

\arrayrulecolor gray!50 \arrayrulecolor black P3

0.930

\pm

0.034

0.5

\pm

0.4

7.7

\pm

5.8

0.841

\pm

0.041

0.4

\pm

0.2

9.9

\pm

6.1

0.688

\pm

0.168

1.0

\pm

1.3

17.3

\pm

12.4

0.648

\pm

0.140

0.8

\pm

0.4

14.1

\pm

5.0

8/10

15/15

\arrayrulecolor gray!50 \arrayrulecolor black P4

0.926

\pm

0.031

0.5

\pm

0.2

7.3

\pm

2.4

0.827

\pm

0.043

0.4

\pm

0.1

9.5

\pm

3.3

0.660

\pm

0.159

1.0

\pm

1.5

19.5

\pm

13.5

0.647

\pm

0.126

0.7

\pm

0.3

13.5

\pm

5.0

8/10

15/15

\arrayrulecolor gray!50 \arrayrulecolor black P5

0.914

\pm

0.034

0.6

\pm

0.3

7.9

\pm

3.2

0.806

\pm

0.043

0.4

\pm

0.2

10.6

\pm

3.4

0.628

\pm

0.170

1.2

\pm

1.8

23.9

\pm

14.9

0.552

\pm

0.186

1.0

\pm

0.5

16.2

\pm

3.4

9/10

15/15

Table 3: Mean and standard deviations of the geometrical metrics (Dice, ASSD and HD) for the participants P1-P6 and the foundation models (mcMedSAM and MedSAM-f) on the MYOSAIQ challenge test set. Bold face numbers are the best results for the automatic methods in each column and blue numbers indicate results that are not statistically significantly different from the best result in each column (paired t-test,

>0.05

). TP represents the number of times a MVO structure was present in an image and correctly detected. TN represents the number of times a method correctly identified that no MVO structure is present in an image. Given that the methods do not segment the same number of patients for MVO, we decided not to provide a ranking for this structure. Inter-obs corresponds to the inter-expert variability computed from two different experts on the same testing dataset.

\Block 2-1 Method

\Block 1-3 LV

\Block 1-3 MYO

\Block 1-3 MI

\Block 1-3 MVO

Dice

ASSD (mm)

HD (mm)

Dice

ASSD (mm)

HD (mm)

Dice

ASSD (mm)

HD (mm)

Dice

ASSD (mm)

HD (mm)

0.928; 0.004 [0.921, 0.935]

0.5; 0.0 [0.4, 0]

7.2; 0.3 [6.7, 7.7]

0.840; 0.004 [0.832, 0.848 ]

0.4; 0.0 [0.3, 0 ]

9.0; 0.3 [8.4, 9.6]

0.669; 0.02 [0.627, 0.706]

1.0; 0.2 [0.7, 0]

19.8; 1.6 [16.8, 23.2]

0.538; 0.063 [0.404, 0.652]

1.2; 0.3 [0.7, 0]

19.6; 4.6 [12.3, 29.5]

\arrayrulecolor gray!50 \arrayrulecolor black P2

0.933; 0.003 [0.926, 0.939]

0.4; 0.0 [0.4, 0]

6.8; 0.2 [6.4, 7.3]

0.840; 0.004 [0.831, 0.848 ]

0.4; 0.0 [0.3, 0 ]

9.0; 0.3 [8.4, 9.7]

0.700; 0.016 [0.668, 0.73]

1.0; 0.2 [0.7, 0]

16.9; 1.5 [14.1, 19.9]

0.681; 0.039 [0.601, 0.756]

0.7; 0.1 [0.5, 0]

14.4; 1.5 [11.6, 17.6]

\arrayrulecolor gray!50 \arrayrulecolor black P3

0.930; 0.004 [0.922, 0.937]

0.5; 0.0 [0.4, 0]

7.7; 0.7 [6.7, 9.2]

0.841; 0.005 [0.832, 0.85 ]

0.4; 0.0 [0.3, 0 ]

9.9; 0.7 [8.7, 11.4]

0.688; 0.019 [0.648, 0.723]

1.0; 0.2 [0.7, 0]

17.3; 1.4 [14.7, 20.2]

0.647; 0.05 [0.545, 0.741]

0.8; 0.2 [0.5, 0]

14.1; 1.8 [10.9, 17.8]

\arrayrulecolor gray!50 \arrayrulecolor black P4

0.926; 0.003 [0.919, 0.933]

0.5; 0.0 [0.4, 0]

7.3; 0.3 [6.8, 7.8]

0.827; 0.005 [0.818, 0.836 ]

0.4; 0.0 [0.4, 0 ]

9.5; 0.4 [8.8, 10.3]

0.659; 0.018 [0.623, 0.693]

1.0; 0.2 [0.7, 0]

19.5; 1.5 [16.6, 22.7]

0.647; 0.045 [0.555, 0.73]

0.7; 0.1 [0.5, 0]

13.5; 1.8 [9.9, 17.0]

\arrayrulecolor gray!50 \arrayrulecolor black P5

0.914; 0.004 [0.906, 0.921]

0.6; 0.0 [0.5, 0]

7.9; 0.4 [7.3, 8.6]

0.806; 0.005 [0.797, 0.816 ]

0.4; 0.0 [0.4, 0 ]

10.6; 0.4 [9.9, 11.4]

0.627; 0.019 [0.59, 0.663]

1.2; 0.2 [0.9, 0]

23.9; 1.7 [20.7, 27.2]

0.553; 0.062 [0.431, 0.674]

1.0; 0.2 [0.7, 0]

16.2; 1.1 [14.0, 18.3]

Table 4: Bootstrap analysis of the segmentation performance on the MYOSAIQ challenge test set. For each participant (P1–P6) and automatic version of the foundation model (mcMedSAM), the Dice, ASSD, and HD metrics are reported as bootstrap mean , standard deviation , and 95% confidence interval (CI), estimated from 5,000 bootstrap resamples of the test subjects. Confidence intervals are reported as [

CI_{\text{lower}}

CI_{\text{upper}}

\Block 2-1 Method	\Block 1-4 LV				\Block 1-4 MYO				\Block 1-4 MI				\Block 1-4 MVO
	CC	MAE (mL)	LOA (mL)	CRPS	CC	MAE (mL)	LOA (mL)	CRPS	CC	MAE (mL)	LOA (mL)	CRPS	CC	MAE (mL)	LOA (mL)	CRPS
P1	0.969	8.2	23.8	0.002	0.967	8.6	24.5	0.004	0.882	3.4	12.1	0.001	0.972	2.7	6.5	0.000
\arrayrulecolor gray!50 \arrayrulecolor black P2	0.980	6.7	18.0	0.011	0.946	11.2	31.1	0.018	0.892	3.4	12.3	0.005	0.989	1.1	3.0	0.003
\arrayrulecolor gray!50 \arrayrulecolor black P3	0.978	7.4	19.6	0.012	0.924	13.1	38.1	0.022	0.888	3.2	11.7	0.006	0.990	2.6	4.9	0.006
\arrayrulecolor gray!50 \arrayrulecolor black P4	0.977	7.5	20.4	0.012	0.918	12.3	38.2	0.021	0.868	3.3	12.6	0.006	0.997	1.4	2.9	0.004
\arrayrulecolor gray!50 \arrayrulecolor black P5	0.976	8.1	19.9	0.014	0.923	17.5	52.5	0.030	0.902	3.5	11.6	0.007	0.970	2.5	5.6	0.006

Table 5: Clinical scores (CC, MAE, LOA, CRPS, see details in section 2.4.2 ) for the participants P1-P6 and the foundation models (mcMedSAM and MedSAM-f) on the MYOSAIQ challenge test set. Bold face numbers indicate the best results for the automatic methods in each column. Given that the methods do not segment the same number of patients for MVO, we decided not to provide a ranking for this structure. Inter-obs corresponds to the inter-expert variability computed from two different experts on the same testing dataset.

Figure 4: Representative example illustrating a typical successful case. P2 and P5 closely match the expert annotation, whereas MedSAM Auto correctly identifies the infarction and microvascular obstruction but with less accurate delineation. Left ventricular cavity (red), healthy myocardium (green), infarction (magenta), and microvascular obstruction (yellow).

Figure 5: Representative challenging case illustrating typical failure modes. P2 and P5 confuse MVO with infarction and miss the expert-annotated infarcted region, whereas MedSAM Auto detects both structures but with localization errors. Left ventricular cavity (red), healthy myocardium (green), infarction (magenta), and microvascular obstruction (yellow).

Figure 6: Distribution of Dice and Hausdorff Distance (HD) scores across participants for each label: LV cavity (red), myocardium (green), MI (magenta) and MVO (yellow). The black solid line indicates the median, while the black diamond represents the mean.

Figure 7: Estimation of lesion (MI) and myocardial (MYO) volumes, along with the distribution of estimation errors across participants. The black line indicates the median, while the black diamond denotes the mean. Paired t-tests were conducted to compare each participant’s volume estimation with the reference derived from expert segmentation. Brackets marked with an asterisk indicate statistically significant differences ( $p<0.05$ ).

Method	LV	MYO	MI	MVO
P2	0.5 $\pm$ 0.2	0.4 $\pm$ 0.1	1.0 $\pm$ 1.7	0.7 $\pm$ 0.3
\arrayrulecolor gray!50 \arrayrulecolor black P5	0.6 $\pm$ 0.2	0.5 $\pm$ 0.2	1.2 $\pm$ 1.8	1.0 $\pm$ 0.5
mcMedSAM auto	0.4 $\pm$ 0.1	0.4 $\pm$ 0.1	1.3 $\pm$ 1.2	1.9 $\pm$ 1.1
Inter-obs	0.3 $\pm$ 0.3	0.2 $\pm$ 0.2	0.2 $\pm$ 0.3	0.1 $\pm$ 0.1

Table 6: ASSD scores on the 3D-IR-GRE sequences from the test dataset (93% of the test data)

Method	LV	MYO	MI	MVO
P2	0.3 $\pm$ 0.2	0.2 $\pm$ 0.1	1.4 $\pm$ 2.1	– –
\arrayrulecolor gray!50 \arrayrulecolor black P5	0.3 $\pm$ 0.2	0.4 $\pm$ 0.1	0.8 $\pm$ 0.4	– –
mcMedSAM auto	0.3 $\pm$ 0.1	0.4 $\pm$ 0.1	2.2 $\pm$ 1.1	– –
Inter-obs	0.2 $\pm$ 0.2	0.2 $\pm$ 0.1	0.1 $\pm$ 0.1	– –

Table 7: ASSD scores on the 2D-PSIR sequences from the test dataset (7% of the test data). No MVO cases were present in this subset.

Explore similar work

Mar 30, 2026

Curriculum-Guided Myocardial Scar Segmentation for Ischemic and Non-ischemic Cardiomyopathy

Identification and quantification of myocardial scar is important for diagnosis and prognosis of cardiovascular diseases. However, reliable scar segmentation from Late Gadolinium Enhancement Cardiac Magnetic Resonance (LGE-CMR) images remains a challenge due to variations in contrast enhancement across patients, suboptimal imaging conditions such as post contrast washout, and inconsistencies in ground truth annotations on diffuse scars caused by inter observer variability. In this work, we propose a curriculum learning-based framework designed to improve segmentation performance under these challenging conditions. The method introduces a progressive training strategy that guides the model from high-confidence, clearly defined scar regions to low confidence or visually ambiguous samples with limited scar burden. By structuring the learning process in this manner, the network develops robustness to uncertain labels and subtle scar appearances that are often underrepresented in conventional training pipelines. Experimental results show that the proposed approach enhances segmentation accuracy and consistency, particularly for cases with minimal or diffuse scar, outperforming standard training baselines. This strategy provides a principled way to leverage imperfect data for improved myocardial scar quantification in clinical applications. Our code is publicly available on GitHub.

Nivetha Jayakumar, Jonathan Pan, Shuo Wang +6

Department of Electrical and Computer Engineering, University of Virginia, USA · Division of Cardiovascular Medicine, University of Virginia Health System, USA · Department of Computer Science, University of Virginia, USA

Medical Image Segmentation Cardiac MRI

Ask PDF

Jun 1, 2026

Personalized 3D Myocardial Infarct Geometry Reconstruction from Cine MRI for Cardiac Digital Twins

Accurate 3D geometric characterization of myocardial infarction (MI) is essential for building cardiac digital twins (CDTs) to precisely simulate infarct-related electrophysiology. Late gadolinium enhancement magnetic resonance imaging (LGE MRI) is the clinical reference for locating MI, yet its reliance on contrast agents restricts use in renally impaired patients and limits longitudinal follow-ups. As an alternative, contrast-free cine MRI visualizes abnormal ventricular wall motion, which is highly indicative of the infarcted area. In this study, we propose a novel explicit geometry-motion embedded model to fully automatically reconstruct personalized, simulation-ready 3D MI geometries directly from multi-view cine MRIs. Specifically, we construct a 4D (3D + t) biventricular mesh to explicitly extract and decouple geometry-aware and motion-aware features. We further design a dual-branch module for adaptive geometry-motion fusion to capture spatiotemporal dependencies for mapping infarcted region. Furthermore, we introduce multi-scale supervision utilizing an AHA-17 segment-guided cross-attention mechanism to steer the prediction, ensuring biophysically consistent reconstruction. Experimental results on 225 cine MRIs demonstrated that the proposed 3D MI reconstruction achieved high performance with an average Dice score of 0.678

\pm

0.011. In the downstream in-silico electrophysiological simulation evaluations, the results were highly consistent with the LGE-derived ground truth, highlighting the great potential of the proposed model for contrast-free scar characterization and seamless integration into CDT modeling. The code will be released publicly upon acceptance of the manuscript for publication.

Yilin Lyu, Mark YY Chan, Ching-Hui Sia +1

Department of Biomedical Engineering, National University of Singapore, Singapore · Department of Medicine, National University of Singapore, Singapore · Department of Cardiology, National University Heart Centre Singapore, Singapore

3D Reconstruction 3D Medical Imaging Magnetic Resonance Imaging 3D Medical Image Segmentation Digital Twins Cardiac MRI

Ask PDF

May 21, 2026

Physiology and Anatomy Aware Inverse Inference of Myocardial Infarction for Cardiac Digital Twin

Accurate localization of myocardial infarction is essential for risk stratification. While LGE-MRI remains the gold standard, it is resource-intensive. Integrating cine MRI with ECG enables a more detailed representation of infarct properties. Existing inverse MI inference methods overlook realistic scar morphology and cardiac repolarization, reducing sensitivity to subtle ECG variations and interpretability of infarct-induced electrophysiological changes. In this paper, we propose a novel framework for noninvasive MI localization using cardiac digital twins. To bridge the domain gap between simulation and reality, we introduce an anatomy-aware stochastic infarct synthesis strategy to synthesize realistic, irregular scars with border zones, mimicking ischemic transmural progression. We then construct a virtual cohort to simulate QRS-T waveforms, capturing both depolarization and repolarization dynamics. Furthermore, we design a Physiology and Anatomy Aware Network (PAA-Net) that jointly encodes 3D myocardial geometry and multi-lead ECGs to infer infarct areas with varying localizations, sizes, spatial extents, and transmuralities. Experimental results demonstrate that our framework significantly outperforms existing methods in inverse inference, achieving Dice scores of 0.7391 and 0.5503 for scar and border zone segmentation, respectively, while further enhancing the interpretability of the ECG-infarct relationship. Our code will be released upon acceptance.

Mengxiao Wang, Yilin Lyu, Julia Camps +6

Shanghai Jiao Tong University, Shanghai, China · National University of Singapore, Singapore · Universitat Pompeu Fabra, Barcelona, Spain

Electrocardiography 3D Medical Image Segmentation Digital Twins Medical Image Localization

Ask PDF