肺部CT密度增高影的半主动阈值分割方法研究

Semi-Automatic Threshold Segmentation Method for Lung CT Patchy Shadows of Increased Density

目的:使用半主动阈值分割方法实现肺CT斑片状密度增高影的精准分割,为量化评估与统计分析提供帮助。方法:回顾性选取2019年4月至2020年2月深圳市第二人民医院被确诊为病毒性肺炎的CT影像上出现斑片状密度增高影的患者26例进行分析。使用Python仿真平台进行CT切片肺实质的提取、局部区域的直方图分析与自适应的阈值生成策略。26例患者的200张切片分别由两名影像科医师独立标注,并计算"人–人"交并比(IOU)、"人–机"IOU进行比较,验证所提出方法的有效性。结果:"人–机"全局平均IOU指标为0.81,95%置信区间(0.78,0.84),"人–人"全局平均IOU指标为0.54,95%置信区间(0.51,0.58)。结论:本研究所设计的半主动阈值分割方法可以鲁棒稳定地提取精细的病灶结构。

Objective A semi-automatic threshold segmentation method was proposed in order to precisely segment patchy shadows of increased density,providing assistance for quantitative evaluation and statistical analysis of lesions.Methods From April 2019 to February 2020,200 slices from 26 cases with viral pneumonia were recruited for analysis.Python platform was employed for pulmonary parenchyma extraction,local-region histogram analysis and adaptive thresholds designation.A total of 200 slices that contained lesions were delineated twice by two experienced radiologists,respectively.The interpersonal IOU and man-machine IOU were calculated for comparative study,the reliability of proposed method was validated consequently.Results The global mean man-machine IOU was 0.81(95%confidence interval:0.78,0.84),the global mean interpersonal IOU was 0.54(95%confidence interval:0.51,0.58).Conclusion Precise lesions can be precisely extracted by proposed semi-automatic segmentation method,laying foundation for precise and personalized treatment.