PERFORMANCE of a machine learning (ML) based model for diagnosing coronary artery disease (CAD) has been investigated, using myocardial perfusion imaging (MPI) single-photon emission CT (SPECT). As CAD is of the most lethal cardiovascular diseases worldwide, recognising risk factors for this condition is crucial. MPI-SPECT imaging offers a promising route of providing functional evaluation of the myocardium and heart arteries non-invasively.
The current study involved evaluation of the performance of different ML models applied to delta, stress, and rest MPI SPECT radiomics for CAD diagnosis and risk classification. Nine classifiers built from three feature selections and nine ML-based algorithms were compared to identify the most accurate model: gradient boosting, extreme gradient boosting, K-nearest neighbour, decision tree, multi-layer perceptron, random forest, logistic regression, support vector machine, and Naive Bayes. Feature selection was based on the methods maximum relevance minimum redundancy, Recursive Feature Elimination using Random Forest Classifier (RF-RFE), and Boruta. In total, 395 participants with suspected CAD underwent a 48-hour rest-stress MPI SPECT. Of these, 78 patients were found normal, and 317 prone to CAD, amongst whom 135, 127, and 55 had low, intermediate, and high risk, respectively.
Left ventricular myocardium was delineated manually on scans to determine desired volumes for investigation, and stress was induced by dobutamine, dipyridamole, and exercise. Clinical variables like family history, age, and gender were retrieved, alongside 118 radiomic features in the scans. Feature extraction was based on the Image Biomarker Standardisation Initiative (IBSI) and assessed with Standardised Environment for Radiomics Analysis (SERA) protocol. Metrics such as area under the receiver operating characteristic curve, specificity, accuracy, and sensitivity were determined to evaluate model performance.
Researchers did acknowledge that this ML aspect of MPI SPECT is observer dependent, error prone, and time-consuming. As a result, automated, objective approaches for measuring cardiac MPI SPECT are in great demand and expected to emerge following this work. The study findings highlighted the potential of ML models for classifying CAD risk using MPI SPECT images. Future studies are urged to include patients with myocardial infarction and CAD-related clinical factors, such as BMI and hyperlipidaemia, to enhance the generalisability of findings.
