Establishment and validation of an artificial intelligence model for ultrasound image quality control in early pregnancy
10.3760/cma.j.cn131148-20250306-00124
- VernacularTitle:早孕期超声图像质量控制人工智能模型的建立与验证
- Author:
Yuting JIANG
1
;
Qiao ZHENG
1
;
Caixin HUANG
1
;
Ting LEI
1
;
Hongning XIE
1
Author Information
1. 中山大学附属第一医院超声医学科，广州 510080
- Publication Type:Journal Article
- Keywords:
Artificial intelligence;
Ultrasonography;
First-trimester;
Standard plane;
Quality control
- From:
Chinese Journal of Ultrasonography
2025;34(7):563-570
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop a deep learning-based artificial intelligence system for assessing image quality in early pregnancy ultrasound，and to evaluate its performance in anatomical structure identification and quality control.Methods:A retrospective study was conducted by collecting 17 910 static ultrasound images of 8 quality-control planes from fetuses at 11 to 13 +6 weeks of gestation who underwent routine first-trimester ultrasound examinations at the First Affiliated Hospital of Sun Yat-sen University from June 2018 to June 2024. The dataset was divided into a training set（12 536 images），a test set（3 582 images），and a validation set（1 792 images）in a 7∶2∶1 ratio to develop a prenatal-screening artificial intelligence system（PSAIS）and to evaluate its performance in the automatic recognition and quality control of standard planes during early pregnancy. The average precision and mean average precision（mAP）were used to measure the model's ability to recognize the anatomical structures on each plane. Intraclass correlation coefficient（ICC）and Kappa statistics were used to assess the consistency between PSAIS and expert-level sonographers in both plane image quality assessment and standardization. The efficiency of PSAIS was also compared to manual quality control. Results:In the test set，the mAP values for recognizing the anatomical structures of the 8 quality-control planes all exceeded 0.800. In the validation set，PSAIS demonstrated moderate to good agreement with two experts in image quality evaluation：the ICC ranged from 0.713 to 0.843 for one expert and 0.678 to 0.788 for the other，while the Kappa values ranged from 0.590 to 0.768 and 0.530 to 0.702，respectively. In terms of plane standardization scoring，PSAIS showed particularly high agreement with expert ratings on the transventricular view（compliance rate 94.6%，Kappa=0.860）and the four-chamber cardiac view with blood flow（compliance rate 94.1%，Kappa=0.778），with agreement above 70% for the remaining planes. Compared with manual quality-control，PSAIS significantly increased processing speed：the total processing time was only 413 seconds，markedly less than the 77 008 seconds and 94 918 seconds required for manual QC（ P<0.001）. Conclusions:The PSAIS system performs well in recognizing and controlling the quality of standard ultrasound planes in early pregnancy，demonstrating high consistency with expert evaluations and significantly improved processing efficiency. It has potential application value in enhancing the quality and efficiency of early pregnancy screening.
