Digital Health Journal Club | The Effectiveness of Quantitative and Qualitative Analysis of Myocardial Perfusion Imaging for the Stenosis Classification of Coronary Artery

Presented by UQ's Queensland Digital Health Centre and Metro North Health's Queensland Digital Academy.

Coronary Artery Diseases (CADs) are significant causes of mortality worldwide. Myocardial Ischemia (MI) is a subset of CADs that occurs when coronary arteries supply insufficient oxygen to heart muscles due to reduced heart blood flow. As a screening procedure considered a non-invasive method and used in many countries, myocardial perfusion imaging (MPI) is a type of computed tomography (CT) in nuclear cardiology to measure how well blood flows through heart muscles or the degree of the stenosis of coronary arteries. However, human assessment of MPI is considered subjective and suffers from the problem of sub-optimal accuracy and re-interpretability. As a result, too many positive patients of MPI were later sent to undergo coronary angiography (CAG) for confirmation.

In this talk, Dr Leelanupab address and focus on the recent update of the comprehensive performance evaluation on quantitative and qualitative analysis of MPI with AI integration. Several machine-learning techniques (non-deep and deep learning methods) have been studied on the private dataset of the Rajavithi Hospital, collected from a 4DM-SPECT (later known as Corridor4DM) application that reconstructs CT images into polar maps in terms of different modalities (i.e., perfusion, severity, defect severity, and defect blackout) and tested in stress and rest stages. Together with data extracted from polar maps, patient characteristics (e.g., diabetes mellitus, hypertension, dyslipidemia) have also been used to train models for quantitative analysis. Examples of models for quantitative analysis are LightGBM, Random Forest, SVM, and XGBoost. Those for qualitative analysis are EfficientNetV2, EfficientNet(V1), VGG19, ResNet50, and DenseNet121. He also mention a web-based application that deploys the best models among studied ones from an offline study for a clinical study.
​
Speaker: Dr Teerapong Leelanupab.

Recording available here.