1488
Native T1 Mapping Radiomics for Classifying Severity of Pulmonary Regurgitation in Patients with Repaired Tetralogy of Fallot1Department of Biomedical Engineering and Environmental Sciences, National Tsing Hua University, Hsinchu, Taiwan, 2Department of Radiology, Kaohsiung Veterans General Hospital, Kaohsiung, Taiwan, 3Department of Electrical Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan, 4Department of Automatic Control Engineering, Feng Chia University, Taichung, Taiwan, 5Department of Pediatrics, Kaohsiung Veterans General Hospital, Kaohsiung, Taiwan, 6Dpartment of Pediatrics, National Yang Ming Chiao Tung University, Hsinchu, Taiwan
Synopsis
Keywords: Myocardium, Radiomics
Motivation: Pulmonary regurgitation (PR) severity is an important prognostic indicator in repaired Tetralogy of Fallot (rTOF) patients. Radiomics analysis may reveal hidden information about cardiomyopathy in these patients.
Goal(s): This study aimed to develop a radiomics-based classification model by native T1 mapping to identify rTOF patients with moderate-to-severe PR and severe PR.
Approach: A total of 623 radiomic features were extracted from native T1 mapping. We used machine learning for feature selection to identify the best radiomic features that maximize the diagnostic value for classifying cardiac diseases.
Results: Optimal performance was achieved in the proposed segmental mid-slice T1 mapping model.
Impact: The segmental mid-slice radiomics of native T1 mapping showed better classification performance than conventional native T1 values in identifying rTOF patients with moderate-to-severe PR and severe PR. The discerned tissue characteristics offered additional physiopathological information beyond native T1 values.
Introduction
Residual pulmonary regurgitation (PR) can result in right ventricular (RV) dilation and dysfunction in repaired tetralogy of Fallot (rTOF) patients, contributing to elevated morbidity and mortality rates [1-3]. Radiomics analysis of cardiovascular magnetic resonance (CMR) images can provide comprehensive characterization and reveal hidden information of cardiomyopathy of a variety of cardiovascular diseases [4-6].This study aimed to develop a radiomics-based classification model by using native T1 mapping to identify rTOF patients with moderate-to-severe PR and severe PR.
Methods
This study included 59 rTOF patients (age=22.7±5.5 years, male/female=35/24). CMR images were acquired in a 3T scanner (Skyra, Siemens). A single-breath-hold ECG-gated Modified Look-Locker inversion recovery (MOLLI) sequence with a single-shot balanced steady-state free procession (bSSFP) readout was acquired in diastole, with scanning parameters of TR/TE=2.57/1.2 ms, flip angle=35∘, and voxel size=1.1×1.1×8 mm3. The manually delineated regions of interest (ROIs) of LV myocardium on the mid-slice were divided into 6 segments, as recommended by the American Heart Association (AHA).One radiomic task was to distinguish rTOF patients with PR≥25% (n=42) from those with PR<25% (n=17). The other radiomic task was to discriminate rTOF patients with PR≥40% (n=31) from those with PR<40% (n=28). Figure 1 illustrates the radiomics workflow. Pyradiomics (version 3.0.1) [7] was used to extract 89 radiomic features from the whole mid-slice and 623 features from segment 7-12 of mid-slice.
We conducted a five-fold outer loop and a three-fold inner loop to establish the classification models. In the classifying moderate-to-severe PR task (PR≥25%), we employed the mutual information (MI) to select the top 15 features in the mid-slice model and ANOVA to select the top 50 features in the segmental mid-slice model. In this task, we used the decision tree (DT) as classification method. In classifying severe PR model (PR≥40%), we used the ANOVA feature selection method to select the top 5 and 30 features in the mid-slice model and the segmental mid-slice model, respectively. The support vector machine (SVM) classification method was used in this task. A p-value<0.05 was considered statistically significant.
Results
Table 1 summarizes the demographics and cardiovascular characteristics of rTOF patients. The PR≥25% group exhibited more dilated RVEDVi, RVMi, and RVSVi than the PR<25% group (all p<0.05). The PR≥40% group exhibited more dilated RVEDVi, RVESVi, and RVMi than the PR<40% group (all p<0.01).Figure 2 shows the values of the LV myocardial native T1 in segment 7-12 and mid-slice. The native T1 value presents no significant difference between PR<25% and PR≥25% groups and between PR<40% and PR≥40% groups. In Figure 3a, the area under curve (AUC) of segmental mid-slice ragiomics model for distinguishing PR≥25% from PR<25% patients was 0.70, which was higher than the mid-slice model, segmental mid-slice T1, mid-slice T1, and CMR indices models (AUC=0.65, 0.52, 0.55, 0.64, respectively). The AUC of discriminating PR≥40% from PR<40% patients was 0.82, which was higher than the other four models (AUC=0.55-0.74) (Figure 3b).
Figure 4 illustrates the values of the top 3 radiomic features in discriminating the severity of PR. Compared to the PR<25% group, the PR≥25% group presented lower size zone non-uniformity normalized (p=0.013), small area emphasis (p=0.052), and small dependence emphasis (p=0.065) in segment 8. Compared to the PR<40% group, the PR≥40% group exhibited decreased inverse variance (p=0.004), increased interquartile range (p=0.062) and robust mean absolute deviation (p=0.072) in segment 11.
Discussion and Conclusion
Compared to conventional native T1 values, the proposed segmental mid-slice native T1 radiomics-based classification model demonstrated high AUC in differentiating PR≥25% rTOF patients from PR<25% patients and in discriminating PR≥40% patients from PR<40% patients.In the classifying moderate-to-severe PR task, the selected top 3 features in segment 8 revealed a more homogeneous and coarser texture pattern in myocardial tissue in PR≥25% group. In the classifying severe PR task, the selected top 3 features in segment 11 demonstrated greater dispersion and variability in the pattern of myocardial fibrosis in the PR≥40% group. A previous study suggested an association between elevated markers of diffuse fibrosis and myocardial dysfunction and severity of PR in rTOF patients [8]. In our study, the radiomic analysis of myocardial native T1 mapping can reveal the different myocardial T1 distribution patterns in rTOF patients with different severity of PR before substantial changes of conventional native T1 values.
In conclusion, the segmental mid-slice radiomic analysis of native T1 mapping demonstrated better classification performance than conventional native T1 values in identifying rTOF patients with moderate-to-severe PR and severe PR.
Acknowledgements
No acknowledgement found.References
[1] Wessel HU, et al. Exercise performance in tetralogy of Fallot after intracardiac repair. J Thorac Cardiovasc Surg. 1980.
[2] Carvalho JS, et al. Exercise capacity after complete repair of tetralogy of Fallot: deleterious effects of residual pulmonary regurgitation. Br Heart J. 1992.
[3] Gatzoulis MA, et al. Risk factors for arrhythmia and sudden cardiac death late after repair of tetralogy of Fallot: a multicentre study. Lancet. 2000.
[4] Raisi-Estabragh Z, et al. Cardiac magnetic resonance radiomics: basic principles and clinical perspectives. Eur Heart J Cardiovasc Imaging. 2020.
[5] Neisius, U.,et al. Radiomic Analysis of Myocardial Native T1 Imaging Discriminates Between Hypertensive Heart Disease and Hypertrophic Cardiomyopathy. JACC Cardiovascular imaging. 2019.
[6] Antonopoulos, A. S., et al. Machine learning of native T1 mapping radiomics for classification of hypertrophic cardiomyopathy phenotypes. Scientific reports. 2021.
[7] van Griethuysen JJM, et al. Computational radiomics system to decode the radiographic phenotype. Cancer Res. 2017.
[8] Riesenkampff, E., et al. Increased left ventricular myocardial extracellular volume is associated with longer cardiopulmonary bypass times, biventricular enlargement and reduced exercise tolerance in children after repair of Tetralogy of Fallot. J Cardiovasc Magn Reason. 2016.
Figures
