Application of intelligent algorithms in Down syndrome screening during second trimester pregnancy

Table 1 Some major fields in the prenatal down syndrome screening and postnatal follow-up database

S/N	Date of receipt	Pregnant mother number	Sample number
Date of birth	Blood sampling date	Expected age at delivery	T21 age riskvalue
T18 age risk	AFP MoM1	hCGb MoM	uE3 MoM
AFP concentration2	hCGb concentration	uE3 concentration	Gestational age
Body weight	Examination unit	Date of last menstruation	Ethnicity/race
Menstrual cycle	History of abnormal pregnancy	Smoking status	CRL3
BPD4	NT5	Fetal chromosomal karyotype	Pregnancy outcomes

¹Multiple of the median.

²Median concentration.

³Crown-rump length.

⁴Biparietal diameter.

⁵Nuchal translucency. AFP: Alpha-fetoprotein; uE3: Unconjugated estriol; hCGb: Human chorionic gonadotropin.

Table 2 Some data samples used for feature selection

ID	Expected age at delivery	AFP concentration	hCGb concentration	uE3 concentration	Body weight	Gestational age	Result
783	19.633	23.22	30.71	5.35	60.7	17.6	1
1504	39.403	44.1	30.41	3.87	62	16.4	1
4338	30.527	23.73	26.29	5.86	65	18.5	1
6050	34.489	11.53	52.42	9.8	52.5	20.6	1
6412	41.966	24.51	49.67	3.91	58	19	1

AFP: Alpha-fetoprotein; uE3: Unconjugated estriol; hCGb: Human chorionic gonadotropin.

Table 3 Experimental results

Algorithm	DR	FPR	Time
CART+AdaBoost	0.2941	0.2857	0:09:54
SVM+SMOTE-Tomek	1	0.0183	0:09:01
CART+AdaBoost+SMOTE-Tomek,maxdepth=2,iter=100	0.9777	0.2337	0:02:14
CART+AdaBoost+SMOTE-Tomek,maxdepth=2,iter=500	0.9909	0.1382	07:18.0
CART+AdaBoost+SMOTE-Tomek,maxdepth=5,iter=100	0.9987	0.0102	0:03:15
CART+AdaBoost+SMOTE-Tomek,maxdepth=5,iter=500	0.9993	0.0028	0:11:37
SVM+EasyEnsemble	1	0.9973	0.00348
SVM+ADASYN	0.9863	0.0932	0:11:05

CART: Classification and regression tree; FPR: False positive rate; DR: Detection rate; SVM: Support vector machine; SMOTE: Synthetic minority over-sampling technique; ADASYN: Adaptive synthetic sampling.

Table 4 Comparison of algorithms when different T21 threshold values were selected

Risk-cutoff value	Age	SVM		Statistical methods
		FPR	DR	FPR	DR
270	Overall	0.009369	0.9295	0.032	0.913
50	Overall	0.003081	0.913	0.007	0.82
150	Overall	0.007039	0.9268	0.019	0.885
250	Overall	0.008197	0.9178	0.03	0.91
500	Overall	0.01775	0.9266	0.053	0.937
750	Overall	0.02299	0.9319	0.072	0.95
1000	Overall	0.03321	0.9305	0.088	0.958

SVM: Support vector machine; FPR: False positive rate; DR: Detection rate.