Atherosclerotic Plaque Pathological Analysis by Unsupervised <inline-formula> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula>-Means Clustering

Abstract

This paper introduced a high-throughput pathological analysis algorithm by using of unsupervised K-means clustering principle and lab color space. The accuracy of this algorithm was verified by comparing with well-established commercially available software. For each type of pathological staining special for atherosclerotic plaque components analysis, accurate pathological analysis results could be obtained by selecting the appropriate cluster classification number (usually 3 to 5, but not limited to 3 to 5). Bland-Altman and linear regression analysis further confirmed that the self-developed algorithm correlated well with the well-established software (correlation coefficient R2 ranged from 0.72 to 0.99). Moreover, the intraand interobserver coefficient of variation were relatively minor, indicating very good reproducibility. So we draw a conclusion that the self-developed algorithm could reduce the human interference factors, improve the efficiency, and be suitable for a large number of analyses of atherosclerotic pathology.

Keywords

• Atherosclerotic plaque
• pathological analysis
• unsupervised K-means clustering
• CIELAB