Radiomics-based analysis by machine learning techniques improves characterization of functionally significant coronary lesions
European Heart Journal(2022)
Abstract
Abstract Background Computed Tomography Coronary Angiography (CTCA) is an effective non-invasive imaging modality for anatomo-functional assessment of coronary artery disease (CAD). Radiomics features have been used for diagnosis or outcome prediction, however, their potential value for characterizing flow limiting coronary lesions has not been explored. Purpose To assess whether application of novel radiomics and machine learning (ML) techniques on CTCA derived datasets improves characterization of functionally significant coronary lesions. Methods Consecutive patients with stable chest pain and intermediate pre-test likelihood for CAD, who underwent CTCA and PET-or SPECT-Myocardial Perfusion Imaging (MPI) respectively, were prospectively evaluated and included in the analysis. PET-MPI was considered abnormal when >1 contiguous segments showed both stress Myocardial Blood Flow ≤2.3mL/g/min and Myocardial Flow Reserve (MFR) ≤2.5 for 15O-water or <1.79 mL/g/min and ≤2.0 for 13N-ammonia respectively. Defect reversibility (DR) was defined as a summed difference score (SDS) between stress and rest images ≥2. CTCA and functional images were fused to assign each myocardial segment to the pertinent coronary territory. Stenosis severity, plaque characteristics and radiomic plaque features were assessed in the total length of the 3 main coronary vessels. In total, 1765 features were extracted from each vessel and a feature reduction and model creation pipeline was constructed [Figure 1]. Two separate datasets: a) coronary stenosis (≥50%) + plaque characteristics and b) coronary stenosis (≥50%) + plaque characteristics + radiomics were formulated and compared in terms of AUCs accordingly. Results A total of 292 coronary vessels (140 with corresponding PET-MPI data and 152 with SPECT MPI data) were analysed. Plaque burden and stenosis severity were the only independent predictors of impaired myocardial perfusion on PET-MPI, with an AUC = 0.749, (95% CI: 0.658–0.826). Stenosis severity, kurtosis, contrast, interquartile range and entropy were predictors of an abnormal PET-MPI result and their combination resulted in an AUC = 0.854, (95% CI: 0.775–0.914). The difference between the 2 models was statistically significant (p-diff: 0.02, 95% CI: 0.0165–0.194). Stenosis severity was the only predictor of a DR on SPECT-MPI, AUC = 0.624 (95% CI: 0.542–0.702). Small Dependence High Gray Level Emphasis, Cluster Prominence, Region Length, wavelet Median and square Median were predictors of a positive SPECT result, with AUC = 0.816, (95% CI: 0.745–0.875). The difference between the two models was statistically significant (p-diff: 0.006, 95% CI: 0.152–0.329) Conclusion Radiomic futures can be combined with anatomical and morphological characteristics of coronary lesions in CTCA imaging and provide valuable complementary information for characterizing functionally significant coronary lesions. Funding Acknowledgement Type of funding sources: Public grant(s) – EU funding. Main funding source(s): This work was supported from European Regional Development Fund, Operational Programme “Competitiveness, Entrepreneurship and Innovation 2014-2022 (EPAnEK)”, titled: The Greek Research Infrastructure for Personalized Medicine (pMED-GR)
MoreTranslated text
Key words
machine learning techniques,machine learning,radiomics-based
AI Read Science
Must-Reading Tree
Example
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined