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
10.3760/cma.j.cn.115807-20220311-00058
- VernacularTitle: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
- Author:
Zheng WAN
1
;
Bing WANG
;
Qinglei HUI
;
Jing YAO
;
Fengxia GONG
;
Chen LI
;
Linlin ZHANG
;
Xin MIAO
;
Lin LIU
;
Kai ZHANG
;
Yanbing JIAN
;
Sisi HUANG
;
Shengwei LAI
;
Wen TIAN
Author Information
1. 中国人民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，北京 100853
- Keywords:
Thyroid nodule;
Artificial intelligence;
Ultrasonography;
Accurate diagnosis
- From:
Chinese Journal of Endocrine Surgery
2022;16(2):185-189
- CountryChina
- Language:Chinese
-
Abstract:
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 0.54.0 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.
