Abstract
Background: Coronary artery brightness (CAB) on echocardiography has been observed during the acute phase of Kawasaki disease (KD), but its clinical relevance remains unclear. This study aimed to quantify CAB and evaluate its clinical significance using unsupervised machine learning (ML).
Methods and Results: Echocardiographic still images from 89 patients with acute-phase KD were analyzed. Pixel values of the coronary arteries (CAs) were extracted and standardized as Z-scores using brightness around the right coronary cusp as a reference. Mean and median pixel intensity (Z-scores) within the coronary artery region were calculated for each major CA branch. Based on these parameters, K-means clustering stratified patients into 2 clusters. Cluster 1 had significantly greater CA diameters and Z-scores in all 3 major coronary branches, with a higher proportion of patients with a maximum CA Z-score ≥2.5. In addition, levels of total bilirubin and pentraxin 3, both known predictors of intravenous immune globulin (IVIG) resistance, were significantly higher in Cluster 1.
Conclusions: Quantitative CAB analysis combined with unsupervised ML effectively stratified KD patients into subgroups with distinct coronary and biomarker profiles. This method may serve as a novel non-invasive tool to evaluate disease severity and predict IVIG resistance in acute-phase KD.
