人工智能技术联合甲状腺超声影像和数据系统(TI-RADS)分级对甲状腺结节良恶性诊断价值的研究

Study on the diagnostic value of benign and malignant thyroid nodules based on artificial intelligence(AI)technology combined with thyroid ultrasound imaging and data system(TI-RADS)grading

目的通过分析比较人工智能(artificial intelligence,AI)超声智能辅助诊断系统、美国放射协会(American College of Radiology,ACR)的甲状腺影像报告和数据系统(thyroid imaging reporting and data system,TI-RADS)分级及AI+ACR TI-RADS联合诊断效能,研究其对良、恶性甲状腺结节的诊断价值和外科治疗的指导意义。方法前瞻性、连续性采集2021年11月1日至2022年2月26日解放军总医院第一医学中心普通外科医学部甲状腺(疝)外科接受手术治疗的甲状腺结节患者349例(605个结节),男95例,女254例,男∶女比为1.00∶2.67,年龄范围为16~78岁,结节最大径范围0.2~5.6 cm。采用SPSS 26.0、R studio软件进行数据处理,分别统计分析AI、ACR TI-RADS分级及联合诊断诊断效能,并行ROC曲线分析。结果联合诊断的AUC值为0.900,大于AI的0.857及ACR TI-RADS的0.788,差异有统计学意义(Z=7.631,均P<0.001)。联合诊断的灵敏度为95.32%,特异度为84.61%,准确度为92.56%,阳性预测值94.69%,阴性预测值86.27%,漏诊率4.68%,误诊率15.38%,均优于其余两种诊断方法,与术后病理结果有极好符合率(Kappa=0.804,P<0.001)。联合诊断识别不同肿瘤最大径的准确度分别为d≤0.5 cm 89.58%,0.5<d≤1.0 cm 96.09%,1.0<d≤2.0 cm 95.45%,2.0<d≤4.0 cm 87.88%,d>4.0 cm 85.71%,均优于其余两种诊断方法。结论AI+ACR TI-RADS联合应用在甲状腺结节性质评估方面具有一定的初筛价值,联合诊断能更有效地判定甲状腺结节的良恶性。

Objective To study the diagnostic value of the artificial intelligence(AI)diagnostic system,ACR TI-RADS classification and AI+ACR TI-RADS combined diagnostic performance in benign and malignant thyroid nodules and its guiding significance for surgical treatment.Methods From Nov.1,2021,to Feb.26,2022,349 patients with 605 thyroid nodules who received surgical treatment in Department of Thyroid(Hernia)Surgery,Department of General Surgery,the First Medical Center of the PLA General Hospital,were selected.There were 95 males and 254 females,male:female=1:2.67,aged 16-78 years,and the nodule diameter was 0.2-5.6 cm.SPSS 26.0 and R studio software were used for data processing.AI diagnostic system,ACR TI-RADS grading and AI+ACR TI-RADS combined diagnostic efficacy were statistically analyzed,respectively.ROC curve analysis was performed in parallel.Results The AUC value of AI+ACR TI-RADS combined diagnosis was 0.900,greater than 0.857 of AI diagnostic system and 0.788 of ACR TI-RADS,and the difference was statistically significant(Z=7.631,both P<0.001).The sensitivity of the combined diagnosis was 95.32%,the specificity was 84.61%,the accuracy was 92.56%,the positive predictive value was 94.69%,the negative predictive value was 86.27%,the missed diagnosis rate was 4.68%,and the misdiagnosis rate was 15.38%,which were better than the other two diagnostic methods.With an excellent coincidence rate with postoperative pathological results(Kappa=0.804,P<0.001).The accuracy of combined diagnosis in identifying the maximum diameter of different tumors was 89.58%for d≤0.5 cm,96.09%for 0.5<d≤1.0 cm,95.45%for 1.0<d≤2.0 cm,87.88%for 2.0<d≤4.0 cm,and 85.71%for d>4.0 cm,better than the other two diagnostic methods.Conclusions The combined application of AI+ACR TI-RADS has a certain primary screening value in evaluating thyroid nodule properties.The combined diagnosis of the two can more effectively determine the benign and malignant thyroid nodules.