机器学习评价超声引导下细针穿刺过程中甲状腺结节位移对其良恶性的诊断价值

Value of thyroid nodules’moving distance during ultrasound-guided fine needle aspiration cytology by machine learning for differential diagnosis between benign and malignant thyroid nodules

目的测量超声引导下细针穿刺细胞学检查(US-GFNAC)过程中的甲状腺结节平均位移,应用机器学习辅助评价平均位移对结节良恶性的鉴别诊断价值。方法收集经病理确诊的甲状腺结节患者145例,包括恶性结节84例,良性结节61例,记录结节的大小、位置、形态、回声、血流信号等超声特征,所有结节均行US-GFNAC,将记录的穿刺视频应用Free-XrosM后处理,测量穿刺过程中结节平均位移,利用支持向量机构建平均位移诊断结节良恶性的模型。随机抽取85%的病例作为训练集(99例)和验证集(25例)验证该模型的有效性和稳定性,余15%的病例(21例)作为测试集验证该模型的预测能力。绘制受试者工作特征曲线分析平均位移诊断结节良恶性的效能。结果甲状腺良恶性结节在大小、位置、回声、血流信号等方面比较差异均无统计学意义。纳入的病例中,19例结节性甲状腺肿、13例甲状腺腺瘤、8例炎性病变和21例皱缩结节的平均位移分别为(1.50±0.51)mm、(1.52±0.50)mm、(1.01±0.55)mm和(5.31±1.30)mm;84例甲状腺乳头状癌的平均位移为(3.10±1.12)mm。应用机器学习辅助评价US-GFNAC过程中结节平均位移鉴别其良恶性具有很好的诊断价值;在测试集中,其诊断甲状腺结节良恶性的敏感性为77.8%,特异性为75.0%,准确率为76.4%,曲线下面积为0.764。结论应用机器学习辅助评价US-GFNAC过程中结节平均位移是一个鉴别结节良恶性的有效指标,有望成为细胞病理学不明确结节的辅助定性指标。

Objective To explore the value of thyroid nodules’mean moving distance(MMD)during ultrasound guided fine needle aspiration cytology(US-G FNAC)by machine learning for differential diagnosis between benign and malignant thyroid nodules.Methods A total of 145 cases of thyroid nodules diagnosed by pathology were collected,including 84 cases of malignant nodules and 61 cases of benign nodules,the ultrasonic features including size,location,shape,echogenicity and blood flow were recorded.All the patients underwent US-G FNAC.The ultrasound video was post-processed by Free-Xros M,the MMD during US-G FNAC was measured.Support vector machine(SVM)was used to build the model for differential diagnosis of benign and malignant thyroid nodules by MMD.85%cases were selected randomly as training set(n=99)and test set(n=25)to verify effectiveness and stability of the model,and the rest 15%cases(n=21)were selected to verify the predictability of the model.Receiver operating characteristic curve was drawn to analyze the efficacy of MMD in the diagnosis of benign and malignant thyroid nodules.Results There were no significant difference in size,location,echogenicity and blood flow between benign and malignant thyroid nodules.The MMD in 19 patients with goiter,13 patients with adenoma,8 patients with inflammation and 21 patients with shrinking nodules were(1.50±0.51)mm,(1.52±0.50)mm,(1.01±0.55)mm and(5.31±1.30)mm,respectively.The MMD in 84 patients with papillary thyroid carcinoma was(3.10±1.12)mm.The MMD during US-G FNAC evaluated by machine learning showed an excellent diagnostic efficacy in benign and malignant nodules.In test set,the sensitivity,specificity,accuracy and area under the curve were 77.8%,75.0%and 76.4%and 0.764,respectively.Conclusion MMD during US-G FNAC evaluated by machine learning is an auxiliary index for diagnosing benign and malignant nodules and is expected to be an effective qualitative index for unclear nodules in cytopathology.