应用决策树模型评估超声造影对不典型子宫腺肌病与子宫肌瘤鉴别诊断的效能分析

Analysis of the efficacy of differential diagnosis of atypical adenomyosis and uterine fibroids through contrast-enhanced ultrasound by decision tree model

目的探讨应用决策树模型评估超声造影对不典型子宫腺肌病与子宫肌瘤的鉴别诊断效能。方法选取分析2021年4月至2022年4月于中国医科大学附属盛京医院行常规超声检查回报难以鉴别不典型子宫腺肌病及子宫肌瘤患者540例,最终纳入超声造影检查患者95例,其中经病理证实子宫腺肌病64例,子宫肌瘤31例。分别统计子宫腺肌病组和子宫肌瘤组定性分析以及时间-强度曲线(TIC)定量分析数据,包括造影剂进入病灶的时相差异性,即造影剂开始进入病灶时间与最终造影剂全部填充病灶的时间差值。首先通过单因素、多因素分析筛选出纳入决策树模型指标,分别建立基于定性分析指标、基于定性与TIC分析决策树模型,进一步评估两者诊断效能。结果单因素分析结果显示定性分析指标中病灶起始增强模式、增强强度、病灶内造影剂分布、造影后病灶边界在两组间差异有统计学意义(均P<0.05);TIC分析指标中造影剂进入时间(AT)、造影剂达峰时间(TTP)、ΔAT、ΔTTP的绝对值(|ΔAT|、|ΔTTP|)在两组间差异有统计学意义(均P<0.05);病灶时相差异性在两组间差异有统计学意义(P<0.05)。多因素分析进一步筛选后纳入模型,最终得到基于定性分析决策树模型准确率和误判率分别为87.40%、(17.90±3.90)%,基于定性与TIC分析决策树模型准确率和误判率分别为90.50%、(21.10±4.20)%;根据两种模型绘制ROC曲线,得到基于定性分析模型与基于定性与TIC分析模型曲线下面积分别为0.915、0.931。结论根据超声造影图像分析建立的决策树模型对于鉴别诊断不典型子宫腺肌病与子宫肌瘤有较高价值。

Objective To evaluate the effectiveness of contrast-enhanced ultrasound in the differential diagnosis of atypical adenomyosis and fibroids using a decision tree model.Methods The data of cases with difficulty in differentiating atypical adenomyosis from fibroids on conventional ultrasound examination at Shengjing Hospital of China Medical University from April 2021 to April 2022 were selected and analyzed.Ninety-five patients with contrast-enhanced ultrasound examination were finally selected,including 64 patients in the pathologically confirmed adenomyosis group and 31 patients in the fibroids group.The data from the qualitative analysis and the quantitative analysis of the time-intensity curve(TIC)curve were collected separately,including the temporal variability of contrast entry into the lesion,i.e.the difference between the time when the contrast agent started to enter the lesion and the time when the contrast agent finally filled the lesion completely.Indicators were first screened for inclusion in the decision tree model by univariate and multifactorial analyses,and decision tree models based on qualitative analysis indicators,and qualitative and TIC-based analyses were developed to further assess the diagnostic efficacy of both models.Results Through the univariate analysis,it showed that the qualitative analysis indicators of lesion onset enhancement pattern,enhancement intensity,intra-lesion contrast distribution,and post-contrast lesion border were of statistical significance(all P<0.05)between the two groups.The differences in contrast arrive time(AT),contrast time to peak(TTP),|ΔAT|,and|ΔTTP|in the TIC curve analysis indexes were statistically significant between the two groups(all P<0.05).The difference in lesion temporal phase variability was statistically significant between the two groups(P<0.05).After further screening by multifactorial analysis,the accuracy and misdiagnosis rates were 87.40%and(17.90±3.90)%in the qualitative analysis-based decision tree model respectively,and 90.50%and(21.10±4.20)%in the qualitative and TIC curve-based analysis decision tree model respectively.The ROC curves were plotted according to the two groups of models,and the areas under the curves were 0.915 and 0.931 respectively.Conclusions A decision tree model based on ultrasonographic image analysis has diagnostic value for the differential diagnosis of atypical adenomyosis and uterine fibroids.