Recent advancements in data science have spurred the development of artificial intelligence (AI) models designed to estimate gestational age (GA) from ultrasound (US) images. This meta-analysis aims to evaluate the accuracy of these AI models in estimating GA, using ultrasound as the gold standard.
A comprehensive literature search was conducted across multiple databases, including PubMed, CINAHL, Wiley Cochrane Library, Scopus, and Web of Science. Studies were selected if they used AI models to estimate GA, with ultrasound as the reference standard. The risk of bias was assessed using the Quality Assessment for Diagnostic Accuracy Studies-2 (QUADAS-2) tool. The mean error in GA estimation was calculated using STATA version-17, and subgroup analyses were conducted based on factors such as the trimester of GA assessment, type of AI model, study design, and external validation.
Out of 1,039 studies screened, 17 met the inclusion criteria, with 10 being included in the meta-analysis. The studies came from a diverse range of geographical and economic backgrounds, with five (29%) from high-income countries (HICs), four (24%) from upper-middle-income countries (UMICs), one (6%) from low-and-middle-income countries (LMICs), and seven (41%) representing data from multiple income regions.
The pooled mean error in GA estimation was 4.32 days (95% CI: 2.82, 5.83; I²: 97.95%) for 2D images (n=6 studies) and 2.55 days (95% CI: -0.13, 5.23; I²: 100%) for blind sweep videos (n=4 studies). Subgroup analysis based on the trimester of GA assessment revealed the following mean errors: 7.00 days (95% CI: 6.08, 7.92) in the first trimester, 2.35 days (95% CI: 1.03, 3.67) in the second trimester, and 4.30 days (95% CI: 4.10, 4.50) in the third trimester. In studies utilizing deep learning for 2D images, convolutional neural networks (CNN) achieved a mean error of 5.11 days (95% CI: 1.85, 8.37), while deep neural networks (DNN) showed a mean error of 5.39 days (95% CI: 5.10, 5.68).
Most studies exhibited an unclear or low risk of bias across various domains, including patient selection, index test, reference standard, flow and timings, and applicability domain.
The results of this meta-analysis suggest that AI models demonstrate good accuracy in estimating GA, with promising potential for pregnancy dating, particularly in resource-limited settings where access to trained professionals may be scarce.
Related topics:
Preventing Uterine Fibroids: Lifestyle Factors and Medical Insights
Couple Sues Fertility Company Over Costly IVF Embryo Destruction
Antibiotics Before Pregnancy: Risks to Fertility and Fetal Health Revealed
