A radiomic model based on cardiac magnetic resonance for the classification of COVID-19-associated myocardial injury

doi:10.3969/j.issn.1674-8115.2026.10.007

Abstract

Abstract:

Objective ·To compare the diagnostic performance of the cardiac magnetic resonance (CMR)-based radiomic model and the extracellular volume fraction (ECV) model in three groups, patients with coronavirus disease 2019 (COVID-19) accompanied by elevated troponin, patients with COVID-19 accompanied by normal troponin, and healthy controls, thereby providing imaging biomarkers for precise clinical stratification. Methods ·This single-center prospective cohort study, conducted from January 2022 to December 2024, enrolled 70 patients with COVID-19 from Shanghai Sixth People's Hospital, Shanghai Jiao Tong University School of Medicine. These patients were stratified into the elevated troponin group (n=24) and the normal troponin group (n=46). Meanwhile, 70 healthy participants were recruited as the healthy control group. The 140 participants were randomly divided into a training set (n=98) and a test set (n=42) at a 7∶3 ratio. All participants underwent 3.0 T CMR scanning. Radiomic features, including first-order, shape, and texture features, were extracted from short-axis late gadolinium enhancement (LGE) images. After feature dimensionality reduction, the most discriminative core features were selected to construct the CMR radiomic model. A control model based on myocardial ECV was also constructed. The diagnostic performance of the two models was validated in the test set. Model performance was comprehensively evaluated using the area under the receiver operator characteristic curve (AUC), accuracy, precision, sensitivity, specificity, and F1 score. The DeLong test was used to compare the AUCs between the two models. Results ·The CMR radiomic model demonstrated significantly superior performance, as measured by AUC, to the ECV control model in both the training and test sets. Specifically, in the training set, the AUC values for the CMR radiomic model versus the ECV control model were 0.944 vs. 0.741 for the elevated troponin group, 0.903 vs. 0.552 for the normal troponin group, and 0.929 vs. 0.684 for the healthy control group. In the test set, the corresponding values were 0.865 vs. 0.708, 0.691 vs. 0.557, and 0.873 vs. 0.667, respectively. Conclusion ·The CMR-based radiomic model can effectively identify COVID-19-associated myocardial injury and provide imaging evidence for the early and precise assessment of such myocardial injury.

Key words: cardiac magnetic resonance (CMR), coronavirus disease 2019 (COVID-19), myocardial injury, radiomics

CLC Number:

R542.2

Xiao Huoyuan, Quyang Danzeng, Zhang Qingchen, Feng Rui, Liu Yuting, Kang Sang, Pan Jingwei. A radiomic model based on cardiac magnetic resonance for the classification of COVID-19-associated myocardial injury[J]. Journal of Shanghai Jiao Tong University (Medical Science), 2026,(10): 1375-1382.

Figures/Tables 7

TrendMD

References 17

[1]	Harrison S L, Buckley B J R, Rivera-Caravaca J M, et al. Cardiovascular risk factors, cardiovascular disease, and COVID-19: an umbrella review of systematic reviews[J]. Eur Heart J Qual Care Clin Outcomes, 2021, 7(4): 330-339.
[2]	Bavishi C, Bonow R O, Trivedi V, et al. Special Article-Acute myocardial injury in patients hospitalized with COVID-19 infection: a review[J]. Prog Cardiovasc Dis, 2020, 63(5): 682-689.
[3]	He J, Zhang B C, Zhou Q, et al. The prognostic value of myocardial injury in COVID-19 patients and associated characteristics[J]. Immun Inflamm Dis, 2021, 9(4): 1358-1369.
[4]	Lockie T, Nagel E, Redwood S, et al. Use of cardiovascular magnetic resonance imaging in acute coronary syndromes[J]. Circulation, 2009, 119(12): 1671-1681.
[5]	Rudski L, Januzzi J L, Rigolin V H, et al. Multimodality imaging in evaluation of cardiovascular complications in patients with COVID-19: JACC scientific expert panel[J]. J Am Coll Cardiol, 2020, 76(11): 1345-1357.
[6]	Phelan D, Kim J H, Elliott M D, et al. Screening of potential cardiac involvement in competitive athletes recovering from COVID-19: an expert consensus statement[J]. JACC Cardiovasc Imaging, 2020, 13(12): 2635-2652.
[7]	Task Force for the management of COVID-19 of the European Society of Cardiology. ESC guidance for the diagnosis and management of cardiovascular disease during the COVID-19 pandemic: part 2-care pathways, treatment, and follow-up[J]. Eur Heart J, 2022, 43(11): 1059-1103.
[8]	Artico J, Shiwani H, Moon J C, et al. Myocardial involvement after hospitalization for COVID-19 complicated by troponin elevation: a prospective, multicenter, observational study[J]. Circulation, 2023, 147(5): 364-374.
[9]	Chen B H, Shi N N, Wu C W, et al. Early cardiac involvement in patients with acute COVID-19 infection identified by multiparametric cardiovascular magnetic resonance imaging[J]. Eur Heart J Cardiovasc Imaging, 2021, 22(8): 844-851.
[10]	Shafiabadi Hassani N, Talakoob H, Karim H, et al. Cardiac magnetic resonance imaging findings in 2954 COVID-19 adult survivors: a comprehensive systematic review[J]. J Magn Reson Imaging, 2022, 55(3): 866-880.
[11]	Tijmes F S, Marschner C, Thavendiranathan P, et al. Magnetic resonance imaging of cardiovascular manifestations following COVID-19[J]. J Magn Reson Imaging, 2023, 58(1): 26-43.
[12]	Neisius U, El-Rewaidy H, Nakamori S, et al. Radiomic analysis of myocardial native T1 imaging discriminates between hypertensive heart disease and hypertrophic cardiomyopathy[J]. JACC Cardiovasc Imaging, 2019, 12(10): 1946-1954.
[13]	Pujadas E R, Raisi-Estabragh Z, Szabo L, et al. Prediction of incident cardiovascular events using machine learning and CMR radiomics[J]. Eur Radiol, 2023, 33(5): 3488-3500.
[14]	Durmaz E S, Karabacak M, Ozkara B B, et al. Radiomics-based machine learning models in STEMI: a promising tool for the prediction of major adverse cardiac events[J]. Eur Radiol, 2023, 33(7): 4611-4620.
[15]	Baessler B, Luecke C, Lurz J, et al. Cardiac MRI and texture analysis of myocardial T1 and T2 maps in myocarditis with acute versus chronic symptoms of heart failure[J]. Radiology, 2019, 292(3): 608-617.
[16]	Baessler B, Luecke C, Lurz J, et al. Cardiac MRI texture analysis of T1 and T2 maps in patients with infarctlike acute myocarditis[J]. Radiology, 2018, 289(2): 357-365.
[17]	van Griethuysen J J M, Fedorov A, Parmar C, et al. Computational radiomics system to decode the radiographic phenotype[J]. Cancer Res, 2017, 77(21): e104-e107.

Item	Elevated troponin group (n=24)	Normal troponin group (n=46)	Healthy control group (n=70)	P value^①	P value^②	P value^③
Age/year	38.0 (29.0, 52.0)	35.5 (26.8, 41.0)	38.0 (28.0, 56.8)	0.294	0.766	0.159
Hematocrit/%	41.3±5.4	42.0±4.9	43.1±5.2	0.599	0.203	0.323
hs-cTnI/(ng·L^-1)	1 366.0 (114.0, 4 707.7)	2.0 (2.0, 3.0)	2.0 (1.9, 2.0)	<0.001	<0.001	0.001
NT-proBNP/(pg·mL^-1)	639.5 (112.7, 2 194.1)	54.8 (36.8, 98.2)	22.9 (22.9, 39.1)	<0.001	<0.001	<0.001
LVEF/%	61.7 (55.0, 63.7)	57.6 (55.0, 63.2)	61.0 (56.1, 65.0)	0.444	0.402	0.121
T1/ms	1 279 (1 225, 1 319)	1 246 (1 219, 1 268)	1 198 (1 159, 1 234)	0.017	<0.001	<0.001
T2/ms	43 (41, 45)	44 (41, 47)	41 (40, 43)	0.774	0.034	0.007
ECV/%	35±4	32±4	30±4	0.035	<0.001	0.011

Item	Elevated troponin group (n=24)	Normal troponin group (n=46)	Healthy control group (n=70)	P value^①	P value^②	P value^③
Age/year	38.0 (29.0, 52.0)	35.5 (26.8, 41.0)	38.0 (28.0, 56.8)	0.294	0.766	0.159
Hematocrit/%	41.3±5.4	42.0±4.9	43.1±5.2	0.599	0.203	0.323
hs-cTnI/(ng·L^-1)	1 366.0 (114.0, 4 707.7)	2.0 (2.0, 3.0)	2.0 (1.9, 2.0)	<0.001	<0.001	0.001
NT-proBNP/(pg·mL^-1)	639.5 (112.7, 2 194.1)	54.8 (36.8, 98.2)	22.9 (22.9, 39.1)	<0.001	<0.001	<0.001
LVEF/%	61.7 (55.0, 63.7)	57.6 (55.0, 63.2)	61.0 (56.1, 65.0)	0.444	0.402	0.121
T1/ms	1 279 (1 225, 1 319)	1 246 (1 219, 1 268)	1 198 (1 159, 1 234)	0.017	<0.001	<0.001
T2/ms	43 (41, 45)	44 (41, 47)	41 (40, 43)	0.774	0.034	0.007
ECV/%	35±4	32±4	30±4	0.035	<0.001	0.011

Item	Training set (n=98)	Test set (n=42)	P value
Age/year	38.0 (28.0, 51.0)	36.5 (26.3, 49.3)	0.896
Hematocrit/%	41.9±5.3	43.2±4.7	0.202
hs-cTnI/(ng·L^-1)	2.0 (2.0, 3.7)	2.0 (1.9, 3.0)	0.567
NT-proBNP/(pg·mL^-1)	39.9 (22.9, 70.5)	39.4 (22.9, 83.4)	0.891
LVEF/%	60.3 (56.0, 64.0)	61.0 (54.0, 64.1)	0.842
T1/ms	1 224 (1 173, 1 268)	1 228 (1 209, 1 275)	0.260
T2/ms	42 (40, 45)	41 (39, 45)	0.180
ECV/%	32±5	32±4	0.894

Item	Training set (n=98)	Test set (n=42)	P value
Age/year	38.0 (28.0, 51.0)	36.5 (26.3, 49.3)	0.896
Hematocrit/%	41.9±5.3	43.2±4.7	0.202
hs-cTnI/(ng·L^-1)	2.0 (2.0, 3.7)	2.0 (1.9, 3.0)	0.567
NT-proBNP/(pg·mL^-1)	39.9 (22.9, 70.5)	39.4 (22.9, 83.4)	0.891
LVEF/%	60.3 (56.0, 64.0)	61.0 (54.0, 64.1)	0.842
T1/ms	1 224 (1 173, 1 268)	1 228 (1 209, 1 275)	0.260
T2/ms	42 (40, 45)	41 (39, 45)	0.180
ECV/%	32±5	32±4	0.894

Feature	Feature class	Filter	Definition
Autocorrelation	GLCM	Wavelet-HHL	Measures the correlation between the intensities of neighboring voxels
JointAverage	GLCM	Wavelet-HHL	Measures the mean of the joint gray-level distribution of neighboring voxel pairs
Strength	NGTDM	Log.sigma.2.0mm	Measures of the strength/distinctiveness of texture primitives
SumAverage	GLCM	Wavelet-HHL	Measures the mean of the sum of gray levels for neighboring intensity pairs
LargeDependenceLowGrayLevelEmphasis	GLDM	Log.sigma.1.0mm	Measures the extent of large, contiguous low-gray-level regions