基于CA125的机器学习在CT预测浆液性卵巢癌腹腔复发中的应用

Application of CA125-based machine learning in CT prediction of abdominal recurrence of serous ovarian cancer

目的利用机器学习,在监测晚期卵巢癌患者糖类抗原(carbohydrate antigen,CA)125水平的基础上,通过CT预测腹部复发。方法回顾分析本院2018年1月至2020年1月接受CT检查的卵巢癌患者78例的临床资料,患者年龄(65.3±3.8)岁。由两位高年资医师阅读患者CT影像评估是否腹腔复发并检测患者血清CA125水平。构建机器学习支持向量机来预测患者腹腔复发,采用COX回归模型分析相关因素与患者腹腔复发的关系。结果最佳拟合模型为线性模型,其CA125绝对值、绝对变化量、相对变化量、相对变化率分别为0.54、0.55、0.57、0.61;多项式模型CA125绝对值、绝对变化量、相对变化量、相对变化率分别为0.53、0.51、0.55、0.59;径向基核模型CA125绝对值、绝对变化量、相对变化量、相对变化率分别为0.49、0.44、0.51、0.53。多因素回归显示年龄(OR=1.48,95%CI:1.35~1.88)、原发肿瘤大小(OR=1.58,95%CI:1.46~1.79)、肿瘤减灭术程度(OR=1.61,95%CI:1.53~1.82)、CA125相对变化率(OR=1.75,95%CI:1.59~1.90)均与患者腹腔复发相关,差异均有统计学意义(均P<0.05)。结论基于CA125的机器学习能较好预测CT监测的浆液性卵巢癌腹腔复发,相关因素按危险程度由高及低依次为CA125相对变化率、肿瘤减灭术程度、原发肿瘤大小、年龄。

Objective To use machine learning to predict abdominal recurrence through CT based on the monitoring of carbohydrate antigen(CA)125 level in patients with advanced ovarian cancer.Methods A retrospective analysis was carried out on the clinical data of 78 ovarian cancer patients aged(65.3±3.8)who underwent CT examination in our hospital from January 2018 to January 2020.Two senior physicians read the patients'CT images to evaluate whether there was abdominal recurrence or not,the patients'serum CA125 level was detected.A machine learning support vector machine was constructed to predict the patients'abdominal recurrence,and the COX regression model was used to analyze the relationship between related factors and the patients'abdominal recurrence.Results The best-fitting model was a linear model,and the absolute value,absolute change,relative change,and relative change rate of CA125 were 0.54,0.55,0.57,and 0.61,respectively;the absolute value,absolute change,relative change,and relative change rate of CA125 of the polynomial model were 0.53,0.51,0.55,and 0.59,respectively;the absolute value,absolute change,relative change,and relative change rate of the radial basis core model CA125 were 0.49,0.44,0.51,and 0.53,respectively.Multivariate regression showed that age(OR=1.48,95%CI:1.35~1.88),primary tumor size(OR=1.58,95%CI:1.46~1.79),degree of tumor reduction surgery(OR=1.61,95%CI:1.53~1.82),and the relative change rate of CA125(OR=1.75,95%CI:1.59~1.90)were all related to the patients'abdominal recurrence,with statistically significant differences(all P<0.05).Conclusion Machine learning based on CA125 can better predict the abdominal recurrence of serous ovarian cancer monitored by CT.The related factors with risk degree from high to low are the relative change rate of CA125,the degree of tumor reduction surgery,the size of primary tumors,and age.