Cine MR-Based Radiomics to Predict Myocardial Segments With Infarction


 Background: Although myocardial infarction (MI) can be assessed quantitatively and qualitatively by using late gadolinium-enhanced (LGE) cardiovascular magnetic resonance (CMR) imaging, intravenous administration of gadolinium can expose patients to high risk of nephrogenic systemic fibrosis, especially in those with cardiovascular diseases. The purpose of this study is to harness cine CMR-based radiomics for predicting MI without introducing gadolinium.Methods: In this retrospective study, we included 48 patients with acute myocardial infarction (AMI) confirmed by later gadolinium enhancement (LGE) at CMR. CMR examinations were performed within 2 to 6 days after PCI. According to the LGE, each myocardial segment was dichotomized into with and without MI. Radiomic features of myocardial segments were extracted from cine CMR images and the myocardial segments were divided into training and validation sets randomly at a ratio of 0.7:0.3. Pearson correlation and Mann-Whitney U rank test were used to eliminate redundant and irrelevant features. A least absolute shrinkage and selection operator (LASSO) algorithm was used for features selection in the training set. Radiomic signatures were constructed in both the training and validation sets and its predictive performance was assessed using area under the cure of receiver operating characteristic (AUC-ROC).Results：Of 768 myocardial segments in the 48 patients, there were 291 (38%) segments with MI and 477 (62%) segments without MI. After univariate analysis, there were 22 RFs related to MI with statistical significance. LASSO regression selected 18 RFs for radiomics signature builting. AUC-ROC of radiomic signatures in prediction of segments with MI was 0.74(95% CI:0.69-0.78)and 0.68 (95%CI: 0.60-0.75) in the training and validation sets, respectively. The difference was not statistically significant (p=0.14).Conclusion: Cine MR-based radiomics signature can achieve a good prediction performance for MI, which showed the potential to be a promising imaging biomarker for MI without the administration of contrast agent.


Indroduction
Emergency percutaneous coronary intervention (PCI) for myocardial salvage is the current standard of treatment in patients with acute ST segment elevation myocardial infarction (MI) [1] . Echocardiography can be used for immediate and readily bedside assessment of ventricular wall motion. Unfortunately, ventricular wall motion at rest cannot be used to quantify salvage early after infarction, as both necrotic and viable myocardium may demonstrate impaired regional myocardial contraction [2] and the severity of MI is a determining factor of left ventricular dysfunction and long-term remodeling [3] . Therefore, it is essential to identify the size of infarcted myocardium. Cardiovascular magnetic resonance (CMR) imaging, which can provide high spatial resolution contracture and functional images of heart, has obtained a widespread clinical acceptance in assessment of heart diseases. It is well known that late gadolinium enhanced (LGE, 10-15 min after contrast injection) CMR can demonstrate infarcted myocardium as an area of persistent enhancement [4][5][6] . The intravenous administration of gadolinium-based contrast agents is not only time-consuming, but also exposes the patients to potential high-risk of fatal adverse events, such as nephrogenic systemic brosis which occurs more common in patients with renal failure [7] , because patients with coronary artery disease have a higher prevalence of coexisting chronic kidney disease than patients without coronary artery disease. In order to overcome these drawbacks of contrast enhanced CMR, researchers tried to seek native quantitative parameters of non-contrast CMR for characterizing myocardium [8][9][10] .. As myocardial edema occurs as early as 15 minutes after coronary occlusion, with a myocardial water content of 3-7% after occlusion that increases to 28% after myocardial reperfusion, T2 mapping can archive good diagnostic accuracy in identifying acute myocardial edema [11][12] . With the development of computer science and arti cial intelligence, radiomics, which harness computer to high-thoroughly extract high-dimensional digital features of medical images, had become a promising discipline to bridge medical imaging and precision medicine. For the cine CMR sequence on the basis of T2 relaxation, we hypothesized that cine CMR images harbor features to identify MI. Perhaps these features are too trivial to be perceived by naked eyes. Therefore, the purpose of this study is to explore the feasibility of cine CMR-based radiomics to predict MI in patients after PCI to restore epicardial coronary blood ow.

Patients
This retrospective study was approved by the institutional review board of our hospital, and the requirement for informed consent was waived due to its retrospective nature. Eighty-ve consecutive patients who underwent CMR within 2-6 days after PCI at Shenzhen People's Hospital from October 2017 to December 2020 were found in the hospital information system (HIS). Inclusion criteria were:(a) PCI was performed within 24 hours of onset of myocardial infarction symptoms or clinical diagnosis of AMI, and occluded vessels were successfully opened; (b) CMR was performed within 2-6 days after surgery; and (c) the CMR examination was completed successfully and the images could be read from Picture Archiving and Communication Systems (PACS). Exclusion criteria were: (a) history of PCI before this time; (b) inadequate quality for image segment; (c) failure of extracting radiomics features; (d) microvascular obstruction identi ed within the infarcted myocardium. Finally, 48 patients were included in this study ( Fig. 1), including 43 males and 5 females with an age range of 28-75 years and an average age of 52.6 years.
Cine CMR and LGE imaging acquisition CMR was performed on 3.0T MR system (Magnetom Skyra; Siemens Healthcare) and 18-channel cardiac phased-control coil, equipped with respiratory gating and ECG vector gating. Steady-state precession fast sequence was used to capture standard left ventricular (LV) short axis and LV two-chamber, three-chamber and four-chamber cine CMR. The parameters were TR 3.8ms, TE 1.4ms, FA 80.FOV 340 mm x 340 mm, matrix 139×208, layer thickness 8 mm, layer spacing 1.6 mm, breath hold scanning at the end of expiration, acquisition time 15 ~ 18s. Late gadolinium enhancement imaging was performed 10-15 min after the intravenous administration of a gadolinium-based contrast agent at a dose of 0.1mmol/kg, using a twodimensional T1-weighted reverse recovery gradient echo sequence. Inversion times were individually selected from 250ms-400ms to optimize the nulling of the unaffected myocardium. The other parameters were TR 7.6ms, TE 1.09ms, FA 20, FOV 340 mm× 340 mm, matrix 128×128, layer thickness 8 mm, layer spacing 1.6 mm.

Image analysis and radiomics features extraction
Two experienced diagnostic radiologists (with more than 5-year-experience in reading CMR images) interpreted CMR images on the PACS (picture archiving and communication system) in consensus. MI was de ned as an area of persistent enhancement at LGE images. Myocardium segments were conducted using the American Heart Association proposed 16-segment model. 768 myocardial segments were obtained and each segment was dichotomized into with and without MI. The cine CMR images were downloaded from PACS and transferred into the ITK-SNAP (http://www.itksnap.org) software. The end-diastolic phase images were used for myocardium segmentation. The region of interest (ROI) was manually delineated around the boundary of the 16 segments of the left myocardium by the other two radiologists with more 5-year-experience in reading CMR images. Radiomic features were extracted by using the open-source software Pyradiomics package. The ow diagram of the patient selection and radiomics features extraction process is displayed in

Statistical analysis
Statistical analysis was performed using R software version 4.0.0 (http://www.r-project.org) and SPSS. A p value of less than 0.05 indicated statistical signi cance. The procession of feature selection and dimension reduction were following: rst, intra-class correlation coe cients (ICCs) was used to excluded features with low reproducibility (ICC < 0.75) and average of the included features of two radiologists' segmentation was calculated; second, Pearson correlation coe cient was used to exclude those features highly correlated to each other (if two features were high correlated [r > 0.8], the one with lower area under the curve [AUC] of receiver operating characteristic [ROC]) was removed); third, Mann-Whitney U rank test was used to select those features which were associated with MI with statistical signi cance. Then, the 768 myocardial segments in the 48 patients were divided into training and test sets at a ratio of 0.7:0.3 randomly. The selected features were normalized separately using min-max normalization to reduce the impact of different scales. A least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation was employed to select the strongest features in the training set. Subsequently, these nally selected features were used to build radiomics signature. Predictive performance of radiomics signature for MI was evaluated in both the training and test sets using AUC of ROC. The AUCs were compared with the DeLong test using the 'pROC' package in R.

Disscusion
In this study, we derived of radiomics signature from cine CMR images to predict MI. Our study showed that the cine CMR-based radiomics signature yielded moderately high predictive performance with AUCs of 0.74 and 0.68 in training and validation sets, respectively. In the training set, radiomics signature of myocardial segments with MI was higher than that of those without MI with statistical signi cance (− 0.19 ± 0.91 vs -0.94 ± 0.80, p < 0.001).
Administration of contrast materials can be fatal to patients with acute myocardial infarction, for they often coexist chronic kidney disease. As myocardial infarction accompanies water content, T2 relaxation might be prolonged. Cine CMR imaging is one of the most common CMR sequences for heart function assessment. It bases on steady state free precession, which is a gradient recall echo sequence addressing T2 relaxatio. Therefore, it can re ect T2 relaxation change. Recently, Bustin and colleagues developed a single-shot T2prepared balanced SSFP sequence to obtain 3D T2 mapping of myocardium [13] . They observed that focal areas of increased T2 values in patients with myocarditis, which well agreed with LGE ndings. To best of our knowledge, this is the rst study to harness routine cine CMR-based radiomics signature to predict MI in patients with AMI. Although the predictive performance was still moderate. Our study con rmed that radiomics can mine high dimension features beyond human eye perceive ability from the conventional non-contrastenhanced images at some special elds in the real world to overcome some clinical dilemma. Among the 18 selected features, GLDM_DependenceNonUniformity, rst order_10Percentile and GLRLM_RunEntropy were the top three features with positive λ coe cient, while GLDM _GrayLevelNonUniformity and First order_90Percentile had the largest two scales with negative λ coe cient. Therefore, a segment with more heterogeneous texture and high signal intensity indicates the possibility of harboring MI, while homogeneous texture and low signal intensity favor a segment without MI. This phenomenon could be interpreted as prolonging T2 relaxation due to edema after MI. If a segment harbored infarcted and normal myocardium, its texture would be more heterogeneous.
There were several limitations of this study. First, this was a retrospective study, which included those patients with LGE proven MI. Therefore, selection bias was introduced inevitably. Second, due of a small group of patients with MI, we performed analysis about per segment instead of regarding patient, so that clinical variables and offensive vascular were not taken into account. Integrating easy available clinical data might improve predictive performance. Third, this primary study was performed at single center with small sample size, further test cohort from broad available institutes will be required to validate the possibility and usefulness of cine CMR-based radiomics signature to be a surrogate biomarker for MI.
In conclusion, the cine CMR-based radiomics signature showed moderately high predictive performance for MI. This primary single center study throw light on the possibility of harnessing routine cine CMR images to build radiomics signature as a surrogate biomarker for MI. The work ow of the radiomics analysis of myocardial segments.