摘要
目的探讨多中心临床研究对全面部图像中黑色素痣自动检测、分割和定量分析算法的有效性。方法自2019年1—6月,上海交通大学医学院附属第九人民医院、上海市奉贤区奉城医院和上海交通大学附属新华医院基于患者的全面部图像,采用Logistic回归和牛顿法对黑色素痣区域进行检测,再采用Python和OpenCV检测病变边缘,计算区域面积。并通过临床医师检测结果来评估算法的有效性。结果该算法在300例患者中检测出黑色素痣1290个,漏诊80个,误诊125个;召回率为93.57%,精确率为90.31%,F值为0.92。Kappa值>0.8,算法检测与医师检测的面积结果比较,其差异无统计学意义(P=0.720,P>0.05)。结论该算法检测与临床医师诊断的一致性较高。该算法在黑色素痣检测、边缘分割和面积测量等方面都取得了较好的效果。
Objective To explore the effectiveness of automatic detection,segmentation and quantitative analysis of melanin nevi in whole-face images in multi-center clinical trials.Methods Logistic regression and Newton method were used to detect the nevus area based on the whole-face images of patients from January to June 2019.Python and OpenCV were used to detect the lesion edge and calculate the area.The effectiveness of the algorithm was evaluated in multi-center clinical trials.Results The algorithm detected 1290 nevi in300 patients,80 missed diagnosis and 125 misdiagnosis.The recall rate was 93.57%,the accuracy rate was 90.31%,and the F value was0.92.Kappa value between two groups was>0.8;Paired t test showed no significant difference in area between two groups(P=0.720,P>0.05).Conclusion Within the limitations of this study,we demonstrate that the algorithm’s detection is highly consistent with the physician’s diagnosis.The algorithm has achieved good results in detection of nevus,edge segmentation and area measurement.
作者
陈伟
陈敏刚
MOOI WEIJUN
陈骁俊
孙梦哲
柴岗
邓丹
张荣
张艳
CHEN Wei;CHEN Min-gang;MOOI WEIJUN;CHEN Xiao-jun;SUN Meng-zhe;CHAI Gang;DENG Dan;ZHANG Rong;ZHANG Yan(Department of Plastic and Reconstruction Surgery,Fengcheng Hospital of Fengxian District of Shanghai,Shanghai 201411,China)
出处
《中国美容整形外科杂志》
CAS
2021年第6期358-361,共4页
Chinese Journal of Aesthetic and Plastic Surgery
基金
上海市科学技术委员会(18441904500,17411952800)。
