Potential use of radiomics analysis of cine-mode cardiac MRI to detect post-infarction lesions in the left ventricular myocardium
- Authors: Maksimova A.S.1, Samatov D.S.2, Merzlikin B.S.2, Shelkovnikova T.A.1, Listratov A.I.3, Zavadovsky K.V.1
-
Affiliations:
- Tomsk National Research Medical Center of the Russian Academy of Sciences
- National Research Tomsk Polytechnic University
- Siberian State Medical University
- Issue: Vol 5, No 4 (2024)
- Pages: 682-694
- Section: Original Study Articles
- Submitted: 27.04.2024
- Accepted: 18.07.2024
- Published: 13.11.2024
- URL: https://jdigitaldiagnostics.com/DD/article/view/630602
- DOI: https://doi.org/10.17816/DD630602
- ID: 630602
Cite item
Abstract
BACKGROUND: The size and location of an infarct lesion and its clear differentiation from normal tissue are important for clinical diagnosis and precision medicine. This paper is based on the study of radiomic attributes for differentiation of infarct and non infarct tissue using non contrast enhanced cine mode cardiac magnetic resonance imaging (MRI) data.
AIM: The aim of the study was to evaluate the potential use and informative value of radiomics analysis to identify post-infarction lesions in the left ventricular myocardium in patients with ischemic cardiomyopathy (ICM) using non-contrast-enhanced cine-mode cardiac MRI.
MATERIALS AND METHODS: Results of contrast-enhanced cardiac MRI were evaluated in 33 patients following surgical treatment for ICM. Texture analysis was performed on 66 lesions in cine-mode cardiac MRI images, and 105 texture attributes were determined for each lesion. Cardiac MRI was performed according to a standard technique using a Vantage Titan 1.5 T MRI scanner (Toshiba). For texture analysis, 3D Slicer version 5.2.2 (Pyradiomics) was used.
RESULTS: During the study, attribute collinearity diagrams were plotted, zero-significance attributes were identified, and attribute significance was determined using a gradient boosting algorithm, and the cumulative significance of attributes was estimated as a function of their total number. By identifying low-significance attributes, the least significant parameters that did not affect the overall significance level were determined. When single-valued attributes were extracted, no corresponding attributes were found. Based on the analysis results, an ROC curve was constructed for Lasso logistic regression (Se=57.14%, Sp=71.43%, AUC=0.76). The main result of this study was to determine radiomic attributes that characterized lesions corresponding to post-infarction cardiosclerosis and intact left ventricular wall based on cine-mode cardiac MRI images.
CONCLUSIONS: This study demonstrated that radiomics analysis of non-contrast-enhanced cine-mode cardiac MRI images is a promising approach to identify lesions corresponding to myocardial infarction and intact wall. This method may potentially be used to identify lesions of post-infarction cardiosclerosis in patients with ICM without contrast enhancement.
Full Text
Background
The incidence of cardiovascular diseases continues to rise each year. Coronary artery disease is the most prevalent cardiovascular complication and the leading cause of death and disability in adults globally [1]. Myocardial infarction (MI), the most common form of coronary artery disease, is characterized by irreversible cardiac muscle necrosis caused by an acute disruption of coronary circulation [2, 3]. The size and location of the lesion, along with its distinction from normal tissue, are critical for accurate clinical diagnosis and treatment planning [4]. MI is often followed by left ventricular (LV) remodeling, a progressive condition that involves changes in LV size and function within hours after the coronary circulation disturbance [5]. Post-ischemic LV remodeling has a complex pathophysiology, involving various ultrastructural, metabolic, and neurotransmitter processes in the affected and surrounding myocardial tissue. Cardiac remodeling is thought to influence the clinical progression of heart failure [6].
Contrast-enhanced cardiac MRI is widely used and is an important tool for assessing the presence, prevalence, and severity of post-infarction changes in the myocardium. It is also employed to assess myocardial viability and LV remodeling. This technique provides a qualitative assessment of MI and detects microvascular obstruction and hyperemia, which are key factors for determining unfavorable remodeling and predicting adverse cardiovascular outcomes [7–9]. However, this technique has several limitations, including a high dependency on subjective physician judgment and intraoperator variability. Additionally, gadolinium-based contrast agents can lead to nephrogenic systemic fibrosis in patients with renal insufficiency [10], a significant concern given the high prevalence of concurrent renal disorders in patients with cardiovascular diseases [11].
To address these challenges, emerging techniques such as radiomics and texture analysis offer promising alternatives for extracting quantitative data from digital medical images. Radiomics enables a reliable assessment of abnormal changes detected in medical imaging by transforming image data into quantitative measures. Previous studies have explored the potential of texture analysis in cardiac MRI images to differentiate between conditionally normal and nonviable myocardial segments [12]. Some investigations have focused on detecting cicatricial changes in the LV myocardium using non-contrast-enhanced cardiac cine-MRI [13]. Given the morphological differences between affected and healthy myocardium, the corresponding texture features of these areas will also differ [14]. It was assumed that subtle differences between nonviable and conditionally normal segments could be detected on cardiac cine-MRI images using radiomics analysis, based on variations in gray level nonuniformity. However, few studies have confirmed this hypothesis [15, 16]. This theory suggests that post-infarction cardiosclerosis areas could be identified using only non-contrast-enhanced cardiac cine-MRI images, reducing the risks associated with gadolinium-based contrast agents and significantly lowering both the cost and time of analysis. Currently, no such studies have been conducted in patients with ischemic cardiomyopathy (ICM).
Aim
To assess the potential and diagnostic value of radiomics analysis in detecting post-infarction lesions in the LV myocardium in patients with ICM using non-contrast-enhanced cardiac cine-MRI.
Materials and methods
Study design
This observational, single-center, retrospective, cross-sectional, single-arm study involved male and female patients aged 52–65 years who underwent surgical treatment for ICM. All patients received a contrast-enhanced cardiac MRI either as part of their clinical care or according to the study protocol.
Eligibility criteria
The study included patients who met the established criteria for ICM [17]:
1) A history of MI
2) Multivessel coronary artery disease, confirmed by invasive coronary angiography
3) Left ventricular ejection fraction (LVEF) of <40%
4) Increased end-systolic volume (ESV) >60 mL/m2
5) Heart failure classified as New York Heart Association (NYHA) class II–IV
Patients with infectious and rheumatic heart diseases, stroke, acute MI, and right ventricular failure were extracted from the study.
The study used contrast-enhanced cardiac MRI images from patients who underwent surgical treatment for ICM between 2019 and 2023.
Study setting
Patients were enrolled at the Research Institute of Cardiology, Tomsk National Research Medical Center, Russian Academy of Sciences.
The study included patients who underwent cardiac MRI with paramagnetic contrast to assess myocardial viability.
Main study outcome
The primary outcome was the difference in radiomic features between intact myocardium and post-infarction cardiosclerosis (PICS) areas on cardiac cine-MRI images.
Outcomes registration
Contrast-enhanced cardiac MRI
The study reviewed patients’ medical records to gather data from cardiac MRIs with paramagnetic contrast agents, performed to assess myocardial viability. ECG- and respiratory-gated MRI scans were conducted according to standard procedures using a Vantage Titan 1.5-T scanner (Toshiba). Short- and long-axis myocardial images were acquired before and after gadolinium-based contrast injection (gadobutrol 0.1–0.15 mmol/kg body weight). The slice thickness was 7–8 mm, and images were acquired using a 256 × 256 matrix. The MRI protocol included T1- and T2-weighted images, fat-suppressed images to assess the myocardium, dynamic SSFP sequences for LV volume and function assessment, and gradient inversion-recovery (GR-IR) sequences to identify areas of abnormal contrast uptake. The inversion time was selected individually for each case (mean TI, 300 ± 10 ms). Abnormal myocardial changes were assessed using a standardized 17-segment system for LV myocardium segmentation. The primary LV parameters were calculated using segment post-processing software (version 2.2, Medviso AB).
Radiomics analysis
Texture analysis was performed using non-contrast-enhanced cardiac cine-MRI images. All images were segmented using 3D slicer software (version 5.2.2), and radiomic features were automatically extracted using the SlicerRadiomics extension (version aa418a5).
The radiomic features of intact myocardium were compared with those of the PICS areas on non-contrast-enhanced cine-MRI images.
Regions of interest (ROIs) were manually selected to assess differences in radiomic features between intact myocardium and PICS areas. The size and position of the ROIs corresponded to the PICS areas and intact myocardium regions based on time-delayed contrast-enhanced MRI images. Initially, ROIs were manually selected on MRI slices along the short axis (in SSFP mode) that matched the PICS areas on post-contrast MRI images. Texture features were then extracted using the PyRadiomics library. The ROI selection process is illustrated in Fig. 1.
Fig. 1. Selection of regions of interest in post-contrast and non-contrast-enhanced cardiac MRI images, short axis view. a: time-delayed contrast-enhanced MRI showing transmural contrast uptake along the LV inferior wall with no signs of damage in the interventricular septum. b: cardiac cine-MRI showing regions of interest in the posterior wall (green), corresponding to a PICS area in the inferior segment at the middle LV level, and in the anteroseptal segment at the middle level (yellow), corresponding to an intact interventricular septum.
Texture analysis was performed on 66 areas from the cardiac cine-MRI images, with 105 texture features calculated for each area. These texture features were categorized as follows:
- First-order features (Energy, Entropy, Range, Kurtosis, etc.)
- 3D shape features (Mesh Volume, Voxel Volume, Sphericity, etc.)
- 2D shape features (Perimeter, Pixel Surface, Elongation, etc.)
- Gray Level Co-occurrence Matrix
- Gray Level Run Length Matrix
- Gray Level Size Zone Matrix
- Neighboring Gray Tone Difference Matrix
- Gray Level Dependence Matrix
Ethical review
The study was conducted in accordance with Good Clinical Practice and the Declaration of Helsinki. All patients provided written informed consent. The study received approval from the Institutional Review Board of the Research Institute of Cardiology, Tomsk National Research Medical Center (Minutes No. 210, dated February 18, 2021).
Statistical analysis
Statistical processing included the following steps: selection of significant texture features, plotting of feature collinearity diagrams, feature selection based on significance, and application of Lasso regression. Features selection was performed using the following Python functions: identify_collinear, identify_zero_importance, identify_low_importance, identify_single_unique, and identify_all. The sample size was not predetermined.
Results
Participants
Study sample characteristics
The study included 33 patients with ICM. The mean age was 58.3 ± 5.7 years, and 94% of the patients were male. All patients had angina pectoris and heart failure, with NYHA class III being the most prevalent (67% and 61%, respectively). Hypertension was present in 85% of the patients, dyslipidemia in 73%, and diabetes mellitus in 24%. The clinical characteristics of these patients are shown in Table 1.
Table 1. Clinical characteristics of the patients | |
Parameter | Value |
Age, years | 58.3 ± 5.7 |
Male, n (%) | 31 (94%) |
BMI, kg/m2 | 27.5 ± 3.9 |
History of hypertension, n (%) | 28 (85%) |
Heart failure NYHA class, n (%): | |
| 0 (0%) |
| 12 (39%) |
| 20 (61%) |
| 0 (0%) |
Angina pectoris NYHA class, n (%): | |
| 1(3%) |
| 10 (30%) |
| 22 (67%) |
| 0 (0%) |
Diabetes mellitus, n (%) | 8 (24%) |
Dyslipidemia, n (%) | 24 (73%) |
Note. BMI, body mass index; NYHA, New York Heart Association classification. |
Contrast-enhanced cardiac MRI
All patients had a LVEF <40% on contrast-enhanced cardiac MRI. The myocardial mass and LV ESV were elevated. Time-delayed contrast-enhanced MRI identified areas of abnormal contrast uptake corresponding to PICS in all patients. Five (15%) patients had thrombotic masses in a thinned LV wall, and 31 (94%) showed evidence of LV spherical remodeling. The contrast-enhanced cardiac MRI findings are provided in Table 2.
Table 2. Findings from contrast-enhanced cardiac magnetic resonance imaging | |
Parameter | Value |
LVEF, % | 31.5 ± 7.5 |
ESV, mL/m2 | 79.7 ± 16.7 |
LVMM, g | 190.8 ± 2.1 |
VMM, g | 140.8 ± 30.05 |
Number of segments with transmurality >50% | 4.4 ± 2.6 |
Ratio of myocardial mass with contrast uptake to LVMM, % | 27.1 ± 6.9 |
Thrombosis, n (%) | 5 (15) |
Note. LVEF, left ventricular ejection fraction; LVMM, left ventricular myocardial mass; VMM, viable myocardial mass. |
Primary results
Data preprocessing
We removed columns and rows with a missing value rate >0.75. For the remaining data, missing values were imputed using the feature means.
Feature collinearity diagrams
The identify_collinear function was used to detect collinear predictors. For each pair of highly correlated features, the function identified which one to remove. In machine learning, strong correlations between features can increase variance and reduce model interpretability. We identified 33 radiomic features with a correlation coefficient of >0.98. Heat maps were used to visually represent feature collinearity, with columns indicating correlated features and rows indicating the features marked for removal (Fig. 2, 3).
Fig. 2. Heat map showing correlations across the dataset.
Fig. 3. Correlation heat map for 33 features with a correlation coefficient >0.98.
Features with zero importance
The identify_zero_importance function was used to identify features with zero importance. Removing these features does not impact diagnostic performance. Additionally, we applied the FeatureSelector function and a gradient boosting algorithm to assess feature importance. To minimize variance, the importance value was averaged over 10 training iterations. Early stopping with a control dataset was used to prevent overtraining. Fig. 4 shows the normalized importance of the most significant features, with the X-axis representing the normalized importance of each feature.
Fig. 4. Normalized importance values.
We also assessed the cumulative importance of the features based on their total number. We found that 27 features contributed to the overall variation (Fig. 5).
Fig. 5. Changes in cumulative feature importance.
Features with low importance
The identification of features with low importance was based on the same approach used previously. The identify_low_importance function was used to identify features with minimal importance that do not affect the overall outcome. We found that 27 features were needed to achieve a total importance of 0.98, whereas 78 features contributed no additional value to the total importance.
Features with a single value
To identify features with a single value, we selected columns containing only one distinct value. These features exhibit zero variance and are not informative for machine learning. Using this method, we found no features with a single unique value (Fig. 6).
Fig. 6. Number of unique values for each feature.
We applied Lasso logistic regression to select features and generate a receiver operating characteristic (ROC) curve (Fig. 7). The training accuracy and test accuracy were 0.77 and 0.64, respectively (sensitivity, 57.14%; specificity, 71.43%).
Fig. 7. ROC curves for training accuracy (AUC 0.77) and test accuracy (AUC 0.64).
Discussion
Main study outcome
This study assessed the potential of radiomics analysis of non-contrast-enhanced cardiac cine-MRI images for detecting areas corresponding to PICS and intact myocardial tissue in patients with ICM. Using Lasso regression, the method achieved a specificity and sensitivity of 57.14% and 71.43%, respectively. The findings support the ability to differentiate between cicatricial changes in the myocardium and conditionally normal tissue. The relatively low sensitivity and specificity are likely attributable to the small sample size.
Discussion of primary results
The study findings suggest that radiomic features extracted from cine-MRI images can help in identifying post-infarction lesions, thereby potentially improving MI detection and reducing the risks associated with gadolinium-based contrast agents. Few studies have focused on texture analysis of non-contrast-enhanced cardiac cine-MRI images, and none were found in patients with ICM.
These findings align with the study by Smith et al., which showed the significance of machine learning-based radiomic features from non-contrast-enhanced cardiac MRI images for distinguishing between MI and normal myocardial tissue, providing new avenues for clinical diagnosis (AUC 0.88) [16]. Similarly, another study showed that radiomics analysis of non-contrast-enhanced cardiac MRI images in patients with ST-elevation MI (STEMI) helps to assess unfavorable LV remodeling, enhancing diagnostic accuracy and prognosis (AUC 0.82) [18]. Additionally, combining native T1 mapping and extracellular volume mapping in cardiac MRI with radiomics analysis improves the prediction of cardiac function recovery and microvascular damage. Ma et al. demonstrated that radiomics analysis of non-contrast-enhanced T1 mapping images aids in diagnosing acute MI and predicting myocardial function recovery [19]. This approach improves the accuracy of detecting microvascular obstruction and enhances the long-term prognosis of myocardial contractility. Additionally, native T1 mapping-based radiomics can predict major adverse cardiovascular events in patients with STEMI, aiding in risk stratification [20]. Chen et al. found that extracellular volume mapping-based texture analysis can differentiate between reversible and irreversible myocardial damage in STEMI patients, helping to predict unfavorable LV remodeling, which has clinical significance (AUC 0.91) [21]. Another study showed that native T1 mapping-based radiomic features can predict the risk of unfavorable LV remodeling in patients with non-ischemic dilated cardiomyopathy (AUC 0.81) [22]. Modern mapping techniques can effectively detect various myocardial disorders, but their availability is currently limited. We propose an alternative method using non-contrast-enhanced cardiac cine-MRI images, without the need for mapping or contrast enhancement, which provides sufficient accuracy (AUC 0.77).
In recent years, MRI has become the gold standard for noninvasive diagnosis and comprehensive assessment of structural changes in the myocardium [23]. In addition to the well-established diagnostic value of dynamic SSFP sequences for assessing LV volume and function, time-delayed contrast-enhanced MRI is a unique tool for detecting and quantifying PICS areas. The lesion area derived from time-delayed contrast-enhanced MRI findings plays a crucial role in predicting LV remodeling [24]. However, the use of contrast agents is limited to specific patient groups. Many post-infarction patients are clinically unstable during the examination and cannot undergo lengthy procedures. Additionally, gadolinium-based contrast agents may cause side effects, such as renal function impairment, in patients with renal insufficiency.
Study limitations
This study has several limitations, including its retrospective design and small sample size. The sample size required to achieve sufficient statistical power was not determined prior to or during the study. As a result, the sample may not be fully representative, limiting the ability to generalize the findings to the broader population of patients with this condition. Moreover, the study did not include a validation sample to assess the diagnostic value of the model. However, despite the small sample, the study successfully identified significant differences between intact tissue and PICS areas using radiomics analysis of cine-MRI images.
Conclusion
Radiomics analysis of non-contrast-enhanced cardiac cine-MRI images can differentiate between PICS areas and viable myocardium. Therefore, this technique could serve as an alternative to time-delayed contrast-enhanced MRI in patients with MI. However, further studies with larger sample sizes and models with stronger prognostic value are needed to identify patients with ICM and support clinical decision-making in their management.
Additional information
Funding source. This study was not supported by any external sources of funding.
Competing interests. The authors declare that they have no competing interests.
Authors’ contribution. All authors made a substantial contribution to the conception of the work, acquisition, analysis, interpretation of data for the work, drafting and revising the work, final approval of the version to be published and agree to be accountable for all aspects of the work. A.S. Maksimova — planned research design, participated in clinical data collection, data analysis and interpretation and original draft preparation; D.S. Samatov, B.S. Merzlikin — erformed data analysis and interpretation and edited the manuscript; T.A. Shelkovnikova — participated in clinical data collection, data analysis and interpretation and edited the manuscript; A.I. Listratov — performed data analysis and interpretation; K.V. Zavadovsky — planned research design, supervised the study, reviewed and edited the manuscript.
About the authors
Aleksandra S. Maksimova
Tomsk National Research Medical Center of the Russian Academy of Sciences
Author for correspondence.
Email: asmaximova@yandex.ru
ORCID iD: 0000-0002-4871-3283
SPIN-code: 2879-9550
MD, Cand. Sci. (Medicine)
Russian Federation, TomskDenis S. Samatov
National Research Tomsk Polytechnic University
Email: denissamatov470@gmail.com
ORCID iD: 0009-0000-1821-323X
Russian Federation, Tomsk
Boris S. Merzlikin
National Research Tomsk Polytechnic University
Email: merzlikin@tpu.ru
ORCID iD: 0000-0001-8545-9491
SPIN-code: 4815-6169
Cand. Sci. (Physics and Mathematics)
Russian Federation, TomskTatiana A. Shelkovnikova
Tomsk National Research Medical Center of the Russian Academy of Sciences
Email: fflly@mail.ru
ORCID iD: 0000-0001-8089-2851
SPIN-code: 1826-7850
MD, Cand. Sci. (Medicine)
Russian Federation, TomskArtem I. Listratov
Siberian State Medical University
Email: listrat312@gmail.com
ORCID iD: 0009-0004-3202-8179
Russian Federation, Tomsk
Konstantin V. Zavadovsky
Tomsk National Research Medical Center of the Russian Academy of Sciences
Email: Konstz@cardio-tomsk.ru
ORCID iD: 0000-0002-1513-8614
SPIN-code: 5081-3495
MD, Dr. Sci. (Medicine)
Russian Federation, TomskReferences
- Shalnova SA, Drapkina OM, Kutsenko VA, et al. Myocardial infarction in the population of some Russian regions and its prognostic value. Russian Journal of Cardiology. 2022;27(6):4952. EDN: OCPROJ doi: 10.15829/1560-4071-2022-4952
- Desai R, Mishra V, Chhina AK, et al. Cardiovascular disease risk factors and outcomes of acute myocardial infarction in young adults: evidence from 2 nationwide cohorts in the United States a decade apart. Curr Probl Cardiol. 2023;48(9):101747. doi: 10.1016/j.cpcardiol.2023.101747
- Martins Marques T, Hausenloy DJ, Sluijter JP, et al. Girao Intercellular communication in the heart: therapeutic opportunities for cardiac ischemia. Trends Mol Med. 2021;27:248–262. doi: 10.1016/j.molmed.2020.10.002
- Schuleri KH, Centola M, Evers KS, et al. Cardiovascular magnetic resonance characterization of peri-infarct zone remodeling following myocardial infarction. J Cardiovasc Magn Reson. 2012;14:24. doi: 10.1186/1532-429X-14-24
- Bodi V, Monmeneu JV, Ortiz Perez JT, et al. Prediction of Reverse Remodeling at Cardiac MR Imaging Soon after First ST-Segment-Elevation Myocardial Infarction: Results of a Large Prospective Registry. Radiology. 2016;278:54–63. doi: 10.1148/radiol.2015142674
- Del Buono MG, Garmendia CM, Seropian IM, et al. Heart Failure After ST-Elevation Myocardial Infarction: Beyond Left Ventricular Adverse Remodeling. Curr Probl Cardiol. 2022;48(8):101215. doi: 10.1016/j.cpcardiol.2022.101215
- Ibanez B, Aletras AH, Arai AE, et al. Cardiac MRI Endpoints in Myocardial Infarction Experimental and Clinical Trials: JACC Scientific Expert Panel. J Am Coll Cardiol. 2019;74(2):238–256. doi: 10.1016/j.jacc.2019.05.024
- Ussov WYu, Babokin VE, Mochula OV, et al. Contrast-enhanced magnetic resonance tomography in patients with myocardial infarction and supraventricular tachyarrhythmias. Siberian Journal of Clinical and Experimental Medicine. 2014;29(4):33–38. EDN: TBFGPX doi: 10.29001/2073-8552-2014-29-4-33-38
- Usov VYu, Vyshlov EV, Mochula OV, et al. Contrast-ehanced MRI in time structure analysis of myocardial damage in acute infarction and early prehospital thrombolytic therapy. Medical Visualization. 2018;(2):56–69. EDN: XMLLXN doi: 10.24835/1607-0763-2018-2-56-69
- Kuo PH, Kanal E, Abu-Alfa AK, Cowper SE Gadolinium-based MR contrast agents and nephrogenic systemic fibrosis. Radiology. 2007;242(3):647–649. doi: 10.1148/radiol.2423061640
- Kim RJ, Wu E, Rafael A, et al. The use of contrast enhanced magnetic resonance imaging to identify reversible myocardial dysfunction. N Engl J Med. 2000;343(20):1445–1453. doi: 10.1056/NEJM200011163432003
- Kotu LP, Engan K, Eftestol T, et al. Segmentation of scarred and non scarred myocardium in LG enhanced CMR images using intensity based textural analysis. Annu Int Conf IEEE Eng Med Biol Soc. 2011:5698–5701. doi: 10.1109/IEMBS.2011.6091379
- Larroza A, Lopez Lereu MP, Monmeneu JV, et al. Texture analysis of cardiac cine magnetic resonance imaging to detect nonviable segments in patients with chronic myocardial infarction. Med Phys. 2018;45(4):1471–1480. doi: 10.1002/mp.12783
- Maksimova AS, Ussov WYu, Shelkovnikova TA, et al. Cardiac MRI Radiomics: review. Siberian Journal of Clinical and Experimental Medicine. 2023;38(3):13–22. EDN: RUADYI doi: 10.29001/2073-8552-2023-39-3-13-22
- Larroza A, Materka A, Lopez Lereu MP, et al. Differentiation between acute and chronic myocardial infarction by means of texture analysis of late gadolinium enhancement and cine cardiac magnetic resonance imaging. Eur J Radiol. 2017;92:78–83. doi: 10.1016/j.ejrad.2017.04.024
- Avard E, Shiri I, Hajianfar G, et al. Non contrast Cine Cardiac Magnetic Resonance image radiomics features and machine learning algorithms for myocardial infarction detection. Comput Biol Med. 2022;141:105145. doi: 10.1016/j.compbiomed
- Felker GM, Shaw LK, O’Connor CM A standardized definition of ischemic cardiomyopathy for use in clinical research. J Am Coll Cardiol. 2002;39(2):210–218. doi: 10.1016/s0735-1097(01)01738-7
- Liu M, Xin A, Chen T, et al. Non contrast cine cardiac magnetic resonance derived radiomics for the prediction of left ventricular adverse remodeling in patients with ST-segment elevation myocardial infarction. Korean J Radiol. 2023;24(9):827–837. doi: 10.3348/kjr.2023.0061
- Ma Q, Ma Y, Yu T, et al. Radiomics of non contrast enhanced T1 mapping: diagnostic and predictive performance for myocardial injury in acute ST-segment-elevation myocardial infarction. Korean J Radiol. 2021;22(4):535–46. doi: 10.3348/kjr.2019.0969
- Ma Q, Ma Y, Wang X, et al. A radiomic nomogram for prediction of major adverse cardiac events in ST-segment elevation myocardial infarction. Eur Radiol. 2021;31(2):1140–1150. doi: 10.1007/s00330-020-07176-y
- Chen BH, An DA, He J, et al. Myocardial extracellular volume fraction radiomics analysis for differentiation of reversible versus irreversible myocardial damage and prediction of left ventricular adverse remodeling after ST-elevation myocardial infarction. Eur Radiol. 2021;31(1):504–514. doi: 10.1007/s00330-020-07117-9
- Chang S, Han K, Kwon Y, et al. T1 Map-based radiomics for prediction of left ventricular reverse remodeling in patients with non ischemic dilated cardiomyopathy. Korean J Radiol. 2023;24:395–405. doi: 10.3348/kjr.2023.0065
- Frederiksen H, Iorgoveanu C, Mahi A. State of the Art and New Advances: Cardiac MRI. New Advances in Magnetic Resonance Imaging. 2023. Available from: http://dx.doi.org/10.5772/intechopen.112413. doi: 10.5772/intechopen.112413
- Bodi V, Monmeneu JV, Ortiz Perez JT, et al. Prediction of Reverse Remodeling at Cardiac MR Imaging Soon after First ST-Segment Elevation Myocardial Infarction: Results of a Large Prospective Registry. Radiology. 2016;278(1):54–63. doi: 10.1148/radiol.2015142674
Supplementary files
