超声人工智能联合TI-RADS分类在甲状腺结节鉴别诊断中的辅助价值

目的:探讨超声人工智能联合美国放射学会甲状腺影像与报告系统(TI-RADS)分类在甲状腺结节良恶性鉴别诊断中的价值。方法:回顾性分析860例(共920个结节)行甲状腺手术的患者,术前均行超声检查,并与术后组织病理学结果对照,比较人工智能、TI-RADS分类及两者联合诊断的效能,采用Kappa检验分析不同诊断方式的一致性。结果:人工智能、TI-RADS及联合检查诊断甲状腺恶性结节的准确率分别为78.80%(725/920)、80.98%(745/920)及85.00%(782/920);敏感度76.36%(252/330)、80.61%(266/330)及86.36%(285/330);特异度分别为80.17%(473/590)、81.19%(479/590)及84.24%(497/590)。ROC曲线分析人工智能、TI-RADS分类及联合诊断甲状腺恶性结节的AUC分别为0.783、0.792及0.853(Z=1.465,P=0.143)。结论:人工智能与TI-RADS分类对甲状腺结节均具有较高的诊断效能,联合诊断能更有效地鉴别甲状腺结节的良恶性。

Assistant value of ultrasound artificial intelligence combined with TI-RADS classification in differential diagnosis of thyroid nodules

Objective:To explore the value of ultrasound artificial intelligence(AI)combined with American College of Radiology(ACR)thyroid imaging reporting and data system(TI-RADS)in the differential diagnosis of benign and malignant thyroid nodules.Methods:A total of 920 thyroid nodules of 860 patients who underwent thyroid surgery were analyzed retrospectively.Their ultrasound images were reclassified by ACR TI-RADS.The results of AI,TI-RADS classification and joint diagnosis were compared with the results of histopathology,and Kappa test was used to analyze the consistency of different diagnosis methods.Results:The accuracy of AI,TI-RADS and combined examination in the diagnosis of thyroid malignancy were 78.80%(725/920),80.98%(745/920)and 85.00%(782/920),the sensitivity were 76.36%(252/330),80.61%(266/330)and 86.36%(285/330),the specificity were 80.17%(473/590),81.19%(479/590)and 84.24%(497/590).In the ROC curve analysis of AI,TI-RADS classification and joint diagnosis,the AUC of thyroid malignant nodule were 0.783,0.792 and 0.853(Z=1.465,P=0.143),respectively.Conclusions:AI and TI-RADS classification examination have higher diagnostic efficiency for thyroid nodules,and the joint diagnosis can be more effective.