基于核图割算法的冠脉光学相干断层图像斑块区域分割

Plaque region segmentation of intracoronary optical cohenrence tomography images based on kernel graph cuts

冠脉光学相干断层成像(OCT)图像斑块区域分割是冠脉斑块识别的前提和基础,对后续斑块特征分析及易损斑块识别,进而实现冠脉疾病的辅助诊断分析具有十分重要的意义。本文提出了一种新的算法,使用K-means算法与图割算法结合,实现了冠脉OCT图像斑块准确的多区域分割——纤维化斑块、钙化斑块和脂质池,并较好地保留了斑块的边界特征信息。本文实验中对20组具有典型斑块特征的冠脉OCT图像进行了分割,通过与医生手动分割结果比较,证明本文方法能准确地分割出斑块区域,且算法具有较好的稳定性。研究结果证明了本文工作能够极大减少医生分割斑块所消耗的时间,避免不同医生之间的主观差异性,或可辅助临床医生对冠心病的诊断与治疗。

The segmentation of the intracoronary optical coherence tomography （OCT） images is the basis of theplaque recognition, and it is important to the following plaque feature analysis, vulnerable plaque recognition and further coronary disease aided diagnosis. This paper proposes an algorithm about multi region plaque segmentation based on kernel graph cuts model that realizes accurate segmentation of fibrous, calcium and lipid pool plaques in coronary OCT image, while boundary information has been well reserved. We segmented 20 coronary images with typical plaques in our experiment, and compared the plaque regions segmented by this algorithm to the plaque regions obtained by doctor＇s manual segmentation. The results showed that our algorithm is accurate to segment the plaque regions. This work has demonstrated that it can be used for reducing doctors＇ working time on segmenting plaque significantly, reduce sub）ectivity and differences between different doctors, assist clinician＇s diagnosis and treatment of coronary artery disease.