Value of artificial intelligence in assisting ultrasound residents training for the identification，measurement and diagnosis of fetal nuchal translucency thickness
10.3760/cma.j.cn131148-20250228-00108
- VernacularTitle:人工智能在辅助培训超声规培医师识别、测量与诊断胎儿颈项透明层厚度中的价值
- Author:
Liqun FENG
1
;
Siying LIANG
;
Rongbo LING
;
Chengcheng WU
;
Naimin SUN
;
Chunya JI
;
Yuanji ZHANG
;
Xin YANG
;
Dong NI
;
Xuedong DENG
;
Linliang YIN
Author Information
1. 南京医科大学附属苏州医院 苏州市立医院超声医学中心，苏州 215002
- Publication Type:Journal Article
- Keywords:
Artificial intelligence;
Ultrasonography;
Residency standardized training;
Nuchal translucency
- From:
Chinese Journal of Ultrasonography
2025;34(7):579-585
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the clinical application value of artificial intelligence（AI）-assisted training in enhancing the accuracy of nuchal translucency（NT）identification，standardization of measurement，and diagnostic efficacy for abnormalities among ultrasound residents.Methods:A retrospective collection of 300 standard fetal NT ultrasound images was conducted at the Center for Medical Ultrasound，Suzhou Hospital Affiliated of Nanjing Medical University from January 2018 to June 2024. The AI model performed NT measurements and diagnoses once. Four sonographers of different seniority levels（including two resident physicians）independently conducted NT measurements and diagnoses twice. Prior to the experiment，the middle-age and resident sonographers had uniformly completed traditional theory training. Following the first independent measurements，the two resident sonographers received additional AI-assisted training，after which all 4 sonographers performed the second independent measurements. A fetal medicine expert evaluated blindly all the results and compared the differences in NT recognition accuracy，measurement standard rate and diagnosis accuracy between the middle-age sonographer（traditional training only）and two resident sonographers（traditional + AI-assisted training）.Results:For the middle-aged sonographer who only received traditional lecture-based training，the accuracy of NT recognition，standardization rate of measurement，or diagnostic accuracy were not significantly improved befroe and after the training，and the diffrence was not statistically significant（ χ2=0.189，1.887，0.326；all P>0.05）. In contrast，the second-year resident（Resident 2）and first-year resident（Resident 1），who received both traditional lecture-based training and AI training，demonstrated some improvements in the accuracy of NT measurement site recognition，though the differences were not statistically significant（ χ2=1.301，2.418；all P>0.05）. However，both residents did significant improvements in the standardization rate of NT measurement（ χ2=25.768，17.035；all P<0.05）. In terms of diagnostic accuracy，Resident 1 did significant improvement（ χ2=10.180， P<0.05），while Resident 2 also did some improvement，though the difference was not statistically significant（ χ2=2.573， P>0.05）. Conclusions:The AI-assisted training system enhances the ability of ultrasound resident sonographers to recognize，measure，and diagnose NT，providing a novel and efficient training model for standardized residency training in ultrasound specialties.
