摘要
目的通过对比三种计算机自动分割算法对不规则脑出血区域分割的准确性,以得出自动识别不规则脑出血区域的最佳算法。方法选取20例不规则脑出血患者的计算机体层摄影(Computed Tomography,CT)头颅平扫图像作为研究对象,分别采用模糊聚类水平集算法(Fuzzy Clustering and Distance Regularized Level Set Evolution,FCRLS)、阈值法、区域生长法对同一图像进行病灶分割,并以人工勾画病灶区图像作为“金标准”,采用戴斯相似性系数(Dice Similarity Coefficient,DSC)、杰卡德指数(Jaccard Index)、相对体积差(Relative Volume Difference,RVD)比较不同算法对病灶分割的精确度和重合度。结果FCRLS、阈值法、区域生长法的Dice系数分别为(0.88±0.04)、(0.76±0.13)、(0.71±0.12);Jaccard系数分别为(0.80±0.07)、(0.63±0.17)、(0.56±0.14);RVD值分别为(0.12±0.13)、(0.42±0.50)、(0.79±0.52)。FCRLS的Dice、Jaccard系数大于灰度阈值法和区域生长法,差异具有统计学意义(P<0.001);FCRLS的RVD值小于灰度阈值法和区域生长法,差异具有统计学意义(P<0.001)。结论FCRLS、阈值法、区域生长法三种算法比较,FCRLS对于不规则脑出血病灶的自动提取的准确性、鲁棒性均最高,可用于临床辅助诊断和后续治疗。
Objective To compare the accuracy of using three automatic segmentation algorithms for irregular cerebral hemorrhage,the best algorithm is obtained.Methods Head CT plain scan images of 20 patients with irregular cerebral hemorrhage were collected for this study.Three different algorithms,including fuzzy clustering and distance regularized level set evolution(FCRLS),threshold method and region growing segmentation were used to segment lesions on the same image.The manual delineation of lesions was used as the“gold standard”.The dice similarity coefficient(DSC),jaccard index and relative volume difference(RVD)were used to measure the accuracy and coincidence of different algorithms for segmentation of lesions.Results The Dice coefficient values of the three segmentation methods were(0.88±0.04),(0.76±0.13),and(0.71±0.12)respectively.The Jaccard coefficients were(0.8±0.07),(0.63±0.17),and(0.56±0.14),respectively;the RVD values were(0.12±0.13),(0.42±0.50),and(0.79±0.52)respectively.The Dice and Jaccard coefficients of FCRLS were higher than the gray threshold method and the region growing method(P<0.001);the RVD value of FCRLS was lower than the Region Growing method(P<0.001).Conclusion Comparing the FCRLS,threshold method and region growth methods,FCRLS has the highest accuracy and robustness in automatic extraction of irregular cerebral hemorrhage lesions,and can be used for clinical auxiliary diagnosis and follow-up treatment.
作者
杜一凡
瞿航
王苇
赵义
DU Yifan;QU Hang;WANG Wei;ZHAO Yi(Graduate School,Dalian Medical University,Dalian Liaoning 116023,China;Department of Radiology,Affiliated Hospital of Yangzhou University,Yangzhou Jiangsu 225009,China)
出处
《中国医疗设备》
2022年第7期64-67,88,共5页
China Medical Devices
基金
江苏省自然科学基金项目(BK2011451)。
关键词
脑出血
计算机自动分割算法
模糊聚类水平集算法
阈值法
区域生长法
cerebral hemorrhage
automatic computerized image segmentation algorithms
fuzzy clustering and distance regularized level set evolution
threshold method
region growing algorithm
