Skip to main content
. 2023 May 19;14:1130139. doi: 10.3389/fendo.2023.1130139

Table 7.

ML applications in fetal malformations research.

Reference ML application Input ML technique Main output
(111) Complication prediction Mobile collected data DF Ac = 87.5%
(112) Complication prediction Computed tomography images LDA Ac = 95.7%, Se = 92.7% and Sp = 98.9% for craniosynostosis
(113) Complication prediction Ultrasound images SVM AUC = 0.89, Ac = 88.63%, Se = 95%, Sp = 82% and +LR = 5.25 for craniosynostosis
(114) Complication differentiation Stereophotogrammetry images PCA Clear differentiation between craniosynostosis and control patients
(115) Data acquisition Ultrasound videos RF Estimation of heart position, orientation, viewing plane and cardiac phase
(116) Data acquisition Electrocardiography recordings ICA and DT Reconstruction of fetal electrocardiogram
(117) Data acquisition Electrocardiography recordings SDAE Reconstruction of fetal electrocardiogram
(118) Data extraction Ultrasound videos SVM Detection of fetal presentation and heartbeat
(119) Data extraction Cardiotocography recordings EMD Extraction of fetal heart rate
(120) Data extraction Electrocardiography recordings CNN and LSTM Extraction of fetal heart rate
(121) Data extraction Electrocardiography recordings CNN and LSTM Extraction of fetal heart rate
(122) Data extraction Doppler ultrasound recordings EMD Extraction of fetal heart rate
(123) Complication prediction Cardiotocography recordings CNN AUC = 97.82%, Ac = 98.34%, Se = 98.22%, Sp = 94.87% and QI = 96.53% for fetal acidemia caused by hypoxia
(124) Decision making support Cardiotocography recordings Infant software Identification of fetal status
(125) Decision making support Cardiotocography recordings PeriCALM software Identification of fetal status
(126) Decision making support Cardiotocography recordings and ultrasound measurements Foetos software Identification of fetal status
(127) Complication prediction Ultrasound measurements NN Ac = 95% for intrauterine growth restriction
(128) Complication prediction Cardiotocography recordings SVM Ac = 78,26%, Se = 0.78 and Sp = 0.79 for intrauterine growth restriction
(129) Complication prediction Ultrasound images ANN Ac = 91-94% for intrauterine growth restriction
(130) Complication prediction Echocardiography images FINE software Se = 98%, Sp = 93%, +LR = 14 and -LR: 0.02 for congenital heart disease
(131) Complication prediction Echocardiography images CON Ac = 99.0%, Se = 75%, Sp = 99.6%, PPV = 99% and NPV = 88.5% for congenital heart disease
(132) Biomarker discovery Transcriptomics data PCA and K-means miR-1647, miR-3064, mirR-3533, miR-6544, miR-6590, miR-6593, miR-6602, miR-6604, miR-6639, miR-6667, miR-6706, miR-6710, miR-1650, miR-1665, miR-6542, miR-6565, miR-6619 and miR-6706 as novel biomarkers for fetal alcohol spectrum disorder
(133) Complication prediction Clinical parameters LR AUC = 0.880, Se = 1.00, Sp = 0.49, PPV = 0.03 and NPV = 1.00 for macrosomia
(134) Complication prediction Electronic health records LSTM Ac = 93.3% for small, appropriate and large for gestational age
(135) Drug teratogenicity prediction Drug databases information t-SNE and GB AUC = 0.8

ML, machine learning; DF, decision forest; LDA, linear discriminant analysis; SVM, support vector machines; PCA, principal component analysis; RF, random forest; ICA, independent component analysis; DT, decision tree; SDAE, stacked denoising autoencoder; EMD, empirical mode decomposition; CNN, convolutional neural networks; LSTM, long short-term memory; NN, neural networks; ANN, artificial neural networks; CON, compound network; LR, logistic regression; t-SNE, t-distributed stochastic neighbor embedding; GB, gradient boosting; Ac, accuracy; Se, sensitivity; Sp, specificity, AUC, area under the receiver operating characteristic curve; +LR, positive likelihood ratio; QI, quality index; -LR, negative likelihood ratio; PPV, positive predictive value; NPV, negative predictive value.