. 2025 Mar 10;22(1):75–82. doi: 10.4274/tjod.galenos.2025.12689

Table 1. Overview of studies on biomarkers and machine learning in Down syndrome screening.

Authors	Methodology	Sample size	Detection rate	False positive rate	Notable findings
He et al.⁽¹⁸⁾	Machine learning model using random forest algorithms	58.972 pregnant women	66.7% (initial), 85.2% (validation)	5%	Model outperforms traditional methods; strong generalizability; potential for improved prenatal care.
Xu et al.⁽²⁵⁾	Combination of ultrasound markers and non-invasive prenatal testing	856 high-risk single pregnancies	9.46% overall	-	High sensitivity (96.72%) and specificity (98.45%) for chromosomal abnormalities; emphasizes combined modalities.
Zhang et al.⁽²⁶⁾	Deep learning model with convolutional neural network on nuchal ultrasonographic images	822 participants	AUC of 0.98 (training), 0.95 (validation)	-	Surpasses traditional screening based on nuchal translucency and maternal age; potential for universal screening.
Sun et al.⁽²⁷⁾	LASSO method for developing a nomogram based on fetal NT thickness and facial markers	624 cases (302 trisomy 21, 322 euploid)	AUC of 0.983 (training), 0.979 (validation)	-	Strong discrimination ability; provides personalized risk assessment for trisomy 21.
Neocleous et al.⁽²⁸⁾	Non-invasive screening for aneuploidy using artificial neural networks	123.329 cases (122.362 euploid, 967 aneuploid)	100% for Trisomy 21, >80% for others	Minimal	Effective non-invasive screening with financial considerations; optimal false positive and high detection rates.
Volk et al.⁽³¹⁾	Transcriptome analysis to identify biomarkers for Trisomy 21	10 Ts21 and 9 normal euploid samples, plus independent validation	AUC=0.97 to 1.00	-	Transcriptome analysis identifies gene profiles for prenatal Trisomy 21 diagnosis, achieving high classification performance (AUC up to 1.00).

AUC: Area under the curve, LASSO: Least absolute shrinkage and selection operator, NT: Nuchal translucency