脂肪抑制T2WI影像组学模型在子宫肌瘤和子宫腺肌瘤中的鉴别价值

目的：采用影像组学技术探讨磁共振脂肪抑制T2WI在子宫肌瘤与子宫腺肌瘤中的鉴别价值。方法：回顾性分析120例于2015年4月至2019年9月在温州医科大学附属第二医院经术后常规病理检查确诊为子宫肌瘤或子宫腺肌瘤患者，并于手术前行盆腔磁共振平扫检查。以7:3的分配比例随机分成训练组84例和验证组36例，训练组用于预测模型的建立，验证组用于预测模型的优化与验证。在脂肪抑制T2WI的图像上对感兴趣区（ROI）进行高通量数据采集、特征提取及降维，通过多元logistic回归构建预测模型，以敏感度、特异度、准确率、接受者操作特性曲线下面积（AUC）这4个参数评价模型的预测效能。结果：共提取出396个特征参数，筛选并纳入了6个鉴别子宫肌瘤和子宫腺肌瘤相关的影像组学特征参数。在训练组中，影像组学特征构建的预测模型对鉴别子宫肌瘤和子宫腺肌瘤的AUC值为0.861，敏感度为84.1%（37/44），特异度为82.5%（33/40），模型准确率为83.3%（70/84）；在验证组中，鉴别子宫肌瘤和子宫腺肌瘤的AUC值为0.913，敏感度为84.2%（16/19），特异度为94.1%（16/17），模型准确率为88.9%（32/36）。结论：基于脂肪抑制T2WI建立的预测模型在训练组和验证组中均取得了良好的预测性能，该预测模型提供了一种更加优越并且无创的子宫肌瘤和子宫腺肌瘤的鉴别工具。

Fat suppression T2WI with radiomics analysis in differential diagnosis of leiomyoma and adenomyoma

Objective: To identify the value of preoperative fat suppression T2WI of pelvic magnetic resonance imaging (MRI) for differential diagnosis of leiomyoma and adenomyoma with radiomics analysis. Methods: This retrospective study included 120 cases of patients with leiomyoma or adenomyoma pathologically confirmed by surgery in the Second Affiliated Hospital of Wenzhou Medical University between April 2015 and September 2019. All patients underwent conventional MRI plain scan of pelvic before surgery. According to the ratio of 7:3, all the patients were randomly divided into training set (n=84) and verification set (n=36). Region of interest (ROI) was manually delineated on fat suppression T2WI, and subsequently the high-throughput data collection, radiomics features extraction and dimensionality reduction were performed. Multivariate logistic regression was used to build up the prediction model of the differential diagnosis of leiomyoma and adenomyoma. Moreover, the sensitivity, specificity, accuracy and the area under the receiver-operating characteristic curve (AUC) were used for evaluating the predictive performance of the model. Results: A total of 396 radiomics features were extracted and 6 radiomics features were included in the prediction model of the differential diagnosis of leiomyoma and adenomyoma. In the prediction model, AUC in the training set was 0.861, with sensitivity being 84.1% (37/44), specificity 82.5% (33/40) and accuracy 83.3% (70/84), while AUC in the verification set was 0.913, with sensitivity being 84.2% (16/19), specificity 94.1% (16/17) and accuracy 88.9% (32/36). Conclusion: The radiomics prediction model based on fat suppression T2WI can well differentiate the leiomyoma from adenomyoma, which potentially provides a favorable and non-invasive method for clinical differential diagnosis.