Application of artificial intelligence in screening of birth defects a scoping review

MINISTRY OF EDUCATION MINISTRY OF HEALTH HANOI MEDICAL UNIVERSITY ♦♦♦ HUY DO DUC APPLICATION OF ARTIFICIAL INTELLIGENCE IN SCREENING OF BIRTH DEFECTS: A SCOPING REVIEW GRADUATION THESIS DOC TOR OF PREVENTIVE MEDICINE 2015-2021 SUPERVISOR: LE MINH GIANG, MD, PhD HANOI-2021 ACKNOWLEDGMENT I would like to offer my thanks to the staffs of Hanoi Medical University School of Preventive Medicine and Public Health, as well as teachers from the Department of Epidemiology for your guidance and support I would like to express my great appreciation to all the individuals and teams below, that without them I would not be able to accomplish my graduation thesis Firstly I would like to express my deepest gratitude to my supervisor Assoc.Prof Le Minh Giang for his guidance Even though he has always been busy, he still willing to give me some of his precious time This has always been very much appreciated The second and third person that I want to say thanks to is A/Prof Nguyen Thi Trang and Ms Le Thi Minh Phuong from Department of Biomedical Genetics for your help in birth defects, especially Ms Le T1Ũ Minh Phuong for the time you spent screening articles with me In addition, because I can always ask you about scoping review, thank you Ms Nguyen Thi Hue from Center for Research and Training on Substance AbuseHIV Last but not least, I would like to give thanks to my family and close friends, for their support and encouragement throughout my study Hanoi May 2021 Do Due Huy COMMITMENT Respectfully addressed to: Board of Hanoi Medical University Board of Preventive Medicine and Public Health School Department of EpidemiologyBoard of Dissertation Assessment My name is Do Due Huy - Student of Hanoi Medical University, course 2015 - 2021 majoring in Preventive Medicine Doctor, hereby declare that: This is a research that I conducted under the scientific guidance of Assoc.Prof Le Minh Giang The data and results presented in the research are completely truthful In addition, die thesis also uses a number of comments, assessments as well as results from other authors, agencies and organizations, all with source annotations clearly stated in the references I will take full responsibility if tliere was any' fraud in the contents of my research Do Due Huy Hanoi May 2021 ABBREVIATIONS AI Artificial Intelligence SVM Support Vector Machine ANN’S Artificial Neural Networks AFP Maternal Serum Alpha-fetoprotein PAPP-A Pregnancy-associated Plasma Protein A NIPT Non-invasive Prenatal Screening TM/ V*: -u TABLE OF CONTENT ABSTRACT 1.3.1 1.3.2 Definition . - 16 1.3.3 Machine learning MATERIALS AND METHODS 17 ••••••»••••••••••••••••••••••••••••••••• ••••••••• 21 3.1 Protocol and registration 21 3.2 Study subjects 21 3.2.1 Inclusion criteria 21 3.2.2 Exclusion criteria 21 3.3 Information sources 22 3.4 Search 22 3.5 Selection of sources of evidence - 23 3.6 Data charting 3.7 Data Items •• •• • ••••••••••• •• •••••••••••• •• • •• • • • ••••••••• •• • •••••••••• • ••••••• • • • • • • • •• •• • •••••••• • •• •• 2.33 Evaluate the effectiveness of artificial intelligence software 19 3.8 Synthesis of results RE SITT 4.1 Results of articles search 4.2 4.3 _ 23 23 24 25 25 4.4 LIST OF TABLES 4.5 Table 2.1 The risk of Down Edward and Palau Syndrome according to maternal age 4.6 4.7 4.8 LIST OF FIGURES 4.9 4.10 TM/ V*: 4Ả VỈX 4.11 ABSTRACT 4.12 Introduction: There are many advantages of artificial intelligence in healthcare setting compared to human clinicians However, the readiness of Al to replace human as a stand alone screening program is remain unknown 4.13 Method: We conducted a scoping review, a structured evidence synthesis describing a broad research field, to summarise knowledge on Al evaluated for birth defects screening and to assess Al’s ability for adoption in birth defects screening in 2010-2020 period Data were collected through PubNied database using a combination of keywords which was agreed by three different reseachers 4.14 Result: There were 11 eligible studies All Al models built in these studies archieved encouraging results However, there are still key evidence gaps that need to be addressed before Al can be rendered more transferable to large- scale screening evaluations 4.15 Conclusion: We found that die published evidence on Al application for birth defects detection was concentrated around model (algorithmic) development, generally independently of real-world clinical or screening evaluation, and overall the evidence does not indicate readiness of Al systems for real-world birth defects screening trials 4.16 Keywords: Artificial Intelligence; Birth Defects; Screening 4.17 4.18 INTRODUCTION There are about 303.000 newborns who died within weeks of birth every year, worldwide, due to congenital abnomalies* In the vs for every 33 babies bom there is with a birth defect*' Birth defects can have lasting effects, with devastating consequences not only for the child, but also for the family, the health system, and society as a whole* Early screening methods for intervention and treatment can be implemented to limit the complications of birth defects in order to improve quality of life for children and families as well as reducing infant mortality due to birth defects3 Therefore, we need to have effective methods to screen, diagnose and predict birth defects so that we can ha\e early interventions 4.19 In recent years, artificial intelligence (Al) has become an important part of our daily life Some applications of artificial intelligence in everyday life include Google Translate" Google Map5 Youtube video proposal software6 .In healthcare, artificial intelligence lias many important applications such as screening for cancer or building predictive models in disease prevention Application of artificial intelligence to build clinical decision support system is of interested to many scientists around the globe3 The advantages of Al includes promoting evidence-based diagnostics, improving diagnostic efficiency, individualizing treatment as well as bringing high economic efficiency9,10 4.20 With these advantages, the application of artificial intelligence in prenatal screening can be an effective tool to help central hospitals to make more accurate diagnosis of prenatal birth defects as well as replacing doctors in lower level hospitals where there are no trained specialists Therefore, it is necessary' to assess the ability' of artificial intelligence systems to screen for birth defects based on existing literatures and identify' gaps in these studies However, there are not many studies mentioned this issue Thus, we conducted tills study with objectives: 4.21 Objective 1: To describe characteristics of current studies in the 2010 - 2020 period on the application of artificial intelligence in screening of birth defects in pregnant women 4.22 defects Objective 2: To assess Al's readiness in large scale screening of birth 4.527 There is an interesting example propose by IBM on a single node on the decision to go surfing with binary decisions (Yes: No: 0) The decision to go or not to go is our predicted outcome, or y-hat Let’s assume that there are three factors influencing vour decision-making: Are the waxes good? (Yes: No: 0) Is the line-up empty? (Yes: No: 0) Has there been a recent shark attack? (Yes: No: 1) 4.528 rhen, let’s assume the following, giving us the following inputs: 4.529 XI = since the waves are pumping 4.530 X2 = since the crowds are out 4.531 X3 I since there hasn't been a recent shark attack 4.532 Now we need to assign some weights to determine importance Larger weights signify that particular variables are of greater importance to the decision or outcome 4.533 W1 = since large swells don’t come around often 4.534 W2 since you’re used to the crowds 4.535 W3 = since you have a fear of sharks 4.536 Finally, we'll also assume a threshold value of which would translate to a bias value of With all the various inputs, we can start to plug in values into the formula to get the desired output 4.537 Y-hat = (1*5) + (0*2) “ (1*4) 3=6 4.538 If we use the activation function from the beginning of this section, we can determine that the output of this node would be since is greater than In this instance, you would go surfing; but if we adjust the weights or the threshold, we can achieve different outcomes from the model When we observe one decision, like in the above example, we can see how a neural network could make increasingly complex decisions depending on the output of previous decisions or layers 4.539 So in neural network model, each node works as one regression model with its own bias and set of incoming features and weights Then individual model/neuron feeds into numerous other individual neurons across all the hidden layers of the model So we end up with models plugged into other models in a wav where the sum is greater than its parts This allows neural network model to implicitly detect complex nonlinear relationships between independent and dependent variables as well as all possible interactions between predictor variables 4.540 In our study on 11 eligible studies used maternal characteristics as independent variables They are epidemiology informations and result from various screening tests, which can raise nonlinear problem when used to identify birth defects Therefore, the ability to include multiple variables into one models without worrying about their relationships of a neural network model comes in handy in such problem 4.541 b Assess Al’s readiness in large scale screening 4.542 In general our eligible studies showed promising performance 4.543 In congenital heart defects group, models archived accuracy ranged from 85% to 99% with sensitivity of 92% and specificity of 99% in reported study Current studies using traditional methods showed less efficacy in screening of congenital heart defects compared to these two A large European collaborative study archived 39% sensitivity at 1820 gestational week screening58 A meta-analysis study conducted by A Sotiriadis et al showed sensitivity of 44% with 5.5% false postitive rate 55 In another meta analysis on heart defects screening using ductus venosus the sensitivity and specificity were 50% and 93% respectively60 4.544 In chromosome disorder group, highest accuracy’was 100% followed by 99.3% and 98.8% Sensitivity was higher than 95% in on studies reported this variable, similar to specificity, higher than 95% in on reported studies The different in efficacy were less clear while comparing our studies’ models to current methods because they already had good performance enough to be used as popular screening methods In a review study published in 2016 Britton D Rink et al showed that double test, a popular screening method for chromosome disorder, which combine pregnancy-associated plasma protein A (PAPP-A) and free beta hCG with nuchal translucency measurement and maternal age archieved 85% sensitivity and 5% false positive rate 6* Double test is often performed in the first trimester of pregnancy while triple test is used during the second trimester The triple test measures serum levels of AFP estriol, and beta-hCG, with a 70% sensitivity and 5% false-positive rate It is complemented in some regions of the United States, as the Quad screen by adding inhibin A to the panel, resulting in an 81% sensitivity and 5% falsepositive rate for detecting Down syndrome62 63 However, in recent years, the discover}’ that there is sufficient fetal cell free (cf) DNA in maternal plasma to detect Down syndrome has led to the rapid growth of commercial screening for Ulis and other types of aneuploidy This is considered to be the most effective and safest testing method available today A meta-analysis on total of 13 studies have now been published using plasma samples taken prior to invasive prenatal diagnosis for high risk of aneuploidy This study yields a sensitivity of 99.3% 96.9% 87.3% for the detection of trisomy 21 trisomy 18 and trisomy 13 respectively, with 98.3% sensitivity in detecting either of them Despite of high efficiency cfDNA screening test is expensive tliat will only become affordable by public health purchasers if costs fall substantially^ In conclusion, models from our eligible studies had better or equal efficacy' compared to current chromosome disorder screening methods, with relatively lower cost than one method that rival them This suggests effective use of Al algorithms as new methods for screening for birth defects Are they ready in large scale screening? 4.545 We found that the published evidence on Al for birth defects detection was concentrated around model (algorithmic) development, generally independently of realworld clinical or screening evaluation, and overall the evidence does not indicate readiness of Al systems for real-world congenital abnormalities screening trials or for stand-alone screen-reading We arrived at that conclusion despite of encouraging results for the performance of the Al models, because there are still key evidence gaps that need to be addressed before Al can be rendered more transferable to large-scale screening evaluations Although there are still much discussions about the legality and acceptance of pregnant women in applying Al in clinical settings, we give three reasons for this conclusion based on the shortcomings of die technical side of the algorithm building process 4.546 4.547 4.548 St at3$ct udy first larger than audicr 10000 4.549 Repres entative of dataset 4.550 Conge nital heart 4.556 Y Hi ndrrk Teder 4.577 Bi ng Feng A c N A b Koivu 4.569 4.599 Chrom osome disorder Ai A ndteas c Neodeous 4.613 A 4.552 Ji anfeng Yang 4.626 Fetal health status 4.634 4.627 khan Akbuhit A X 4.560 X 4.554 dataset 4.561 X X IT 4.572 X 4.573 X 4.574 X 4.575 4.579 X 4.580 X 4.581 X 4.582 4.587 X 4.588 X 4.589 X 4.590 4.595 X Yes 4.597 4.602 X 4.603 X 4.604 4.609 X 4.610 X 4.611 4.616 X 4.617 X 4.618 4.623 X 4.624 X 4.625 4.630 X 4.631 X 4.632 Y 4.593 4.594 X X 4.600 Y 4.601 X 4.608 X Y 4.615 X 4.621 4.622 X X 4.628 X 4.559 same 4.567 ES Neodeous from X ES 4.614 methods on the 4.566 4.578 4.607 data Al vs X S 4.586 current screening 4.565 ES ndrtas c 4.620 4.571 using testing IT X da Catic 4.606 4.564 4.558 X X 4.584 4.585 4.592 Y S uxh Gong 4.570 4.557 4.553 l different sources ES nhcng Luo 4.563 diseases Ya Mode Bias handling samples 4.555 4.551 4.629 X 4.596 4.568 YES X X X X X X X X X 4.633Table 5.1 Limitation in eligible studies 4.635 Firstly, all studies did not mention the representative of datasets used to build algorithms Most studies had datasets that are collected through screening of pregnant women in only one specific area, leading to homogeneity of datasets and ignoring randomization of groups' characteristics thereby reducing the representativeness of the dataset In addition, ethnic orgin variation in dataset cannot be archived using this sampling method Different ethnic group shows different biochemical marker characteristics 65 So Al algorithms won’t be as effective when meeting different groups of people, which we often encounter in clinical practice, if it was built solely on a single group This problem may be magnified by the small sizes of datasets with less than 10000 samples in on 11 studies So a large and diverse sets of data needs to be considered while building an artificial intelligence algorithm in screening of birth defects Not only that, majority of studies did not undertake validation of file developed Al model using an independent external dataset An independent external dataset allows the encounter of new populations where the proportion of defects is far different than the original one with new features, which push out more mistakes This leads to doubts about models* reported performances This limitation suggests that larger validation datasets, preferably in diverse screening environments and population, are required in order for promising Al algorithm to progress to the next step of clinical development There was one study that collected data from two different countries which reported different ethnic in the population However, its total dataset combined of training and testing set was small < 1S50 individuals) so we can’t conclude that it is representative 4.636 Secondly, no authors wondered whether there was bias in their datasets and described how they handled it We talked about representative based on volumn of the datasets in our first argument We can deal with that problem by collecting more data from other sources However, if we collect data carelessly and let bias in our studies, it can distort our sample’s characteristics further from die original population, which makes the models even worse While most of our eligible studies used collected datasets which prone to selection bias when they did not report including all available observations in their study in case of small dataset or randomly selected observations in case of large database, file collected datasets itself can contain many other bias that even the authors of Al development studies don’t even know about This can create meta bias which leads to very' poor performance of Al models when it comes to clinical setting, bias that authors cannot find out during development process This problem again demonstrates the importance of having an independent external testing dataset Therefore, we suggest that authors should be aware of bias in their datasets, especially in Al related studies where data is the most important factor to be considered and try to explain the way they used to mitigate their inpacts to the result of study 4.637 In non-perfccl contexts where A! program can't archive 100% accuracy in real life practice, then its performance can be inconsistent, just like human It can be good when there is familiar data nor bad if there were new circumstances The results showed above are tempting but we don’t know if the testing data were too easy to classify so that traditional method could achieve such efficacy Comparison between the new Al model and current methods is a must and this must be conducted on the same dataset Otherwise there would be no evidence for Al’s superiority over traditional methods Despite of these arguments tliere were only stud}' reported performance of Al pregram compared to human interpreter Yuxin Gong and colleagues tried to compare their program with expert cardiologists, and die machine was defeated by doctors with senior title, even though it was built using big dataset consist of more than 100.000 samples with accuracy of 85% Future studies should compare Al algorithms with current screening methods in unselected screening examinations, or report the incremental improvement for Al algorithms in combination with other methods It may be that Al algorithms are detecting different findings than current screening methods, and vice versa, but this cannot be determined from the currently available studies This is the final reason for our conclusion 4.638 Our scoping review lias some limitations We searched for articles in only one database, so we couldn't cover all the studies related to Al application in prenatal screening of birth defects In addition, we focused on published studies from 2010 onwards to factor advances in Al capabilities, therefore we did not review older studies that paved the way for more recent Al studies Not only that, we aimed to clarify if Al was ready for prenatal screening application, rather tlian describing the high detailed techniques of Al models And finally, we excluded studies that apply Al in postnatal screening of congenital abnormalities, although postnatal screening also plays an important role in improving individual's health 4.639 CONCLUSION Characteristic of eligible studies 4.640 We found 11 eligible studies All of them were published in the second half of the last decade, more than one third of them were published in China, all were retrospective studies They aimed to develop Al models for prenatal screening of congenital heart defects, chromosome disorders at)d fetus health Al's readiness in large scale screening 4.641 Our scoping review on Al application in prenatal screening of birth defects showed many methodological limitations that present built models from being accepted as a stand-alone method in clinical practice setting Despite of novel techniques and encouraging accuracy, these studies still lack representation in their data and proves for their superior performance against current screening methods Therefore, the methodologic issues highlighted in our work can inform future studies and improve the translation of Al systems into birth defects screening practice 4.642 RECOMMENDATIONS 4.643 According to the results above, we have some recommendation for future studies on Al application in screening of birth defects: - There should be scoping reviews which cover more studies from more databases or explore a broader range of Al application in screening of birth defects, like postnatal screening, or explaining the development of Al methods for more effective Al models - New studies on Al program development should focus more on the representation of training as well as testing datasets alongside comparison with current screening methods while listing protential bias and describe a strategy to deal with them 4.644 REFERENCES Congenital anomalies Accessed November 29, 2020 https://www.who.mt/news-room/fact- sheets/detail/congenital-anomalies coc What are Birth Defects? I CDC Centers for Disease Control and Prevention Published 4.645 4.646 4.647 https://www.forbes.com/sites/bernardmarr/2019/08/23/the-amazing-ways-youtube-uses- artificial-intelligence-and- machine-learning/ Noorbakhsh-Sabet N, Zand R, Zhang Y, Abedi V Artificial Intelligence Transforms the Future of Healthcare Am Med 2019,132(7):795 801 doi:10 1016/j.am|med.2019 01.017 Patel VI, Shortliffe EH, Stefanelh M, el al The Coming of Age of Artificial Intelligence in Medicine Aftif Intell Med 2009;46(l):517 doi:10.1016/j.artmed.2008.07.017 Dilsưian SE, Siegel EL Artificial Intelligence in Medicine and Cardiac Imaging: Harnessing Big Data and Advanced Computing to Provide Personalized Medical Diagnosis and Treatment Curr Cordiol Rep 2013;16(l):441 doi:10.1007/sl 1886-013-0441-8 10 Jha s, Topol EJ Adapting to Artificial Intelligence: Radiologists and Pathologists as Information Specialists JAMA 2O16;316(22):23S3-23S4 doi:10.1001/jama.2016.17438 11 McKenzie JE, 8eller EM, Forbes AB Introduction to systematic reviews and meta-analysis: Introduction to meta-analysis Wolfe R, Abramson M, eds Respifology 2016;21(4):626-637 doi:10.1111/resp.l2783 12 Munn z, Peters MDJ, Stern c, Tufanaru c, McArthur A, Aromataris E Systematic review or scoping review? Guidance for authors when choosing between systematic or scoping review approach BMC Med Res Methodol 2018;18(l):143 doi:10.1186/sl2874-018-0611-x 13 Colquhoun Hl, levac 0, O'Brien KK, Ct al Scoping reviews; time for clarity in definition, methods, and reporting J Clin Epidemiol 2014;67(12):1291-1294 doi:10.1016/j,jclinepi.2014.03.013 14 Pham MT, Rajii A, Greig JD, Sargeant JM, Papadopoulos A, McEwen SA A scoping review of scoping reviews: advancing the approach and enhancing the consistency Res Synth Methods 2014;S(4):371-385 doi:10.1002/jrsm.ll23 15 levac D, Colquhoun H, O'Brien KK Scoping studies: advancing the methodology Implementation Science 2O1O;5(1):69 doi:10.1186/1748-5908-S-69 16 Scoping studies: towards a methodological framework: International Journal of Social 4.648 Research Methodology: Vol 8, No Accessed November 29, 4.649 2020 https://www.tandfonline.com/doi/ab5/10.1080/13645S7032000119616 17 Guidance for conducting systematic scoping reviews: JBI Evidence Implementation 4.650 4.651 Accessed November 29, 2020 https://journals.lww.com/ijebh/Fulltext/2015/09000/Guidance_for_conducting_systematic _scoping_reviews.5.aspx 18 Balancing the evidence: incorporating the synthesis of qualitative data into systematic reviews - Pearson - 2004 - JBI Reports Wiley Online Library Accessed May 23, 2021 https://onlinelibrary.wiley com/doi/abs/10.Ill 1/j 1479-6988 2004.00008.x 19 Trinh Vản Bão Di Truyèn Trường dại hoc Y Hã NÔI 20 King.l Controlling Birth Defects: Reducing the Hidden Toll of Dying and Disabled Children in Low-Income Countries Published online 2008 21 McIntosh GC, Olshan AF, Baird PA Paternal age and the risk of birth defects in offspring Epidemiology 1995;6(3):282-288 doi:10.1097/00001648 199505000 00016 22 Joint WHO-March of Dimes Meeting on Management of Birth Defects and Haemoglobin Disorders (2nd : 2006: Geneva s Organization WH, Dimes M of Management of Birth Defects ond Hoemoglobin Disorders: Report of o Joint WHO March of Dimes Meeting, Geneva, Switzerland, 17-19 May 2006 World Health Organization; 2006 Accessed November 29,2020 https://apps.who.int/iris/handle/10665/43587 23 Iodine in Pregnancy: Is Salt Iodization Enough? I The Journal of Clinical Endocrinology & 4.652 Metabolism I Oxford Academic Accessed November 29, 4.653 2020 https://academic.oup.eom/jcem/article/93/7/2466/2598357 24 Molloy AM, Kirke PN, Brody LC, Scott JM, Mills JL Effects of folate and vitamin B12 deficiencies during pregnancy on fetal, infant, and child development Food Nutr Bull 2008;29(2 Suppl):S101111; discussion SI 12-1 IS doi:10.1177/1564826S080292S114 25 Ngo AD, Taylor R, Roberts CL, Nguyen TV Association between Agent Orange and birth defects: systematic review and metaanalysis Int J Epidemiol 2006;35(5): 1220-1230 dol:10.1093/ije/dyl038 26 Nguyến Ló, Nguyên Viõi Nhản, Cao Ngqc Thành Mót So’Nhte'm Trùng Ảnh Hưởng TỚI ĩhoi Kỳ Nhâ xuãt Đai hoe Huế; 2013 27 Astley $J, 8ailey 0, Talbot c, Clarren SK Fetal alcohol syndrome (FAS) primary prevention through fas diagnosis: II A comprehensive profile of 80 birth mothers of children with FAS Alcohol Alcohol 2000;35(5):509-519 doi:10.1093/alcalc/35.5.509 28 Callahan A Shah NH Chapter 19 - Machine Learning in Healthcare In: Sheikh A, Cresswell KM, Wright A, Bates DW, eds Key Advances in Clinical Informatics Academic Press; 2017:279-291 doi:10.1016/B978-0-12-809523-2.00019-4 29 Tncco AC Lillie E, Zarin w, et al PRISMA Extension for Scoping Reviews (PRlSMA-ScR): Checklist and Explanation Ann Intern Med 2018;169(7):467-473 doi:10.7326/M18-0850 30 Neocleous AC, Nicolaides KH, Schizas CN First Trimester Nonmvastve Prenatal Diagnosis: A Computational Intelligence Approach IEEE J Biomcd Health Inform 2016;20(5):1427-1438 doi:10.1109/JBHI.2015.2462744 31 Luo Y, Li z, Guo H, et al Predicting congenital heart defects: A comparison of three data mining methods PLoS One 2O17;12(5):eO177811 doi:10.1371/journal pone.0177811 32 Neocleous AC, Nicolaides KH, Schizas CN Intelligent Noninvasive Diagnosis of Aneuploidy: Raw Values and Highly Imbalanced Dataset IEEE J Biomed Heolth Inform 2O17;21(5):1271- 1279 doi:10.1109/JBHI.2016 2608859 33 Feng B, Hoskins w, Zhang Y, et al 8i-stream CNN Down Syndrome screening model based on genotyping array BMC Med Genomics 2018;ll(Suppl S):105 doi:10.1186/sl2920-018- 0416-0 34 Neocleous AC, Syngelaki A, Nicolaides KH, Schizas CN Two-Stage approach for risk estimation of fetal trisomy 21 and other ancuploidles using computational intelligence systems UltrasoundObstet Gynecol 2018;51(4):503-508 doi:10.1002/uog.l7558 35 Koivu A, Korpimaki T, Kivela p, Pahikkala T, Sairanen M Evaluation of machine learning algorithms for improved risk assessment for Down's syndrome Comput Biol Med 2018;98:1-7 doi:10.1016/j.compblomed.2018.05.004 36 Akbulut A, Ertugrul E, Topcu V Fetal health status prediction based on maternal clinical history using machine learning techniques Comput Methods Programs Biomed 2O18;163:87-1OO dOKlO.lOie/j.cmpb^OlS.Oe.OlO 37 Application of Neural Networks for classification of Patau, Edwards, Down, Turner and Klinefelter Syndrome based on first trimester maternal serum screening data, ultrasonographic findings and patient demographics - PubMed Accessed May 8, 2021 https://pubmedncbi.nlm.mh.gov/29439729/ 38 Yang J, Ding X, Zhu w Improving the calling of non-invasive prenatal testing on 13-/18-/21- trisomy by support vector machine discrimination PLoS One 2O18;13(12):eO2O784O doi:10.1371/journal.pone.0207840 39 Tcdcr H, Paluoja p, Rckkcr K, Salumcts A, Krjutikov K, Palta p Computational framework for targeted high-coverage sequencing based NIPT PLoS One 2019;14(7):e0209139 doi:10.1371/journal.pone.0209139 40 Gong Y, Zhang Y, Zhu H, Ct al Fetal Congenital Heart Disease Echocardiogram Screening Based on DGACNN: Adversarial One-Class Classification Combined with Video Transfer Learning IEEE Trans Med Imaging 2020;39(4):1206-1222 doi:10.1109/TMI.2019.2946059 41 Oprescu AM, Miró-amarante G, Garcia-Diaz L, Đeltrán LM, Rey VE, Romero-Ternero Me Artificial Intelligence in Pregnancy: A Scoping Review IEEE Access 2020;8:181450-181434 doi: 10.11O9/ACCESS.2O20.3028333 42 The use of machine learning in rare diseases: a scoping review I SpnngerLmk Accessed May 21, 2021 https://link.springer.com/article/10.1186/sl3023-020 01424-6 43 Murphy K Di Ruggiero E, Upshur R, Ct al Artificial intelligence for good health: a scoping review of the ethics literature BMC Medical Ethics 2021:22(1): 14 doi:10.1186/sl2910-021- 00577-8 44 Milunsky A, Milunsky JM Genetic Disorders and the Fetus: Diagnosis, Prevention, and Treatment John Wiley & Sons; 201S 45 Pregnancy Risk Assessment Monitoring System I CDC Published May 3, 2021 Accessed May 24, 2021 https://www.cdc.gov/prams/index.htm 46 Screening tests for you and your baby (STFYAYB) GOV.UK Accessed May 24, 2021 hups://www.gov.uk/government/publication$/screening-tests-for-you-and-your-baby 47 Pregnancy screening National Services Scotland Accessed May 24, 4.654 2021 https://www.nss.nhs.scot/specialist-healthcare/screeningprogrammes/pregnancy- screening/ 48 Sun R, Liu M, lu L, Zheng Y, Zhang p Congenital Heart Disease: Causes, Diagnosis Symptoms, and Treatments Cell Biochcm Biophys 2O15;72(3):857-86O doi:10.1007/sl2013-01S-0551-6 49 What are common treatments for Down syndrome? https://www.nichd.nih.gov/ https://www.nichd.nih.gov/health/topics/down/condĩtioninfo/treatments Accessed May 24, 2021 50 Potts M China's one child policy BMJ 20O6;333(7564):361-362 doi:10.1136/bmj.38938.412S93.80 51 Dai L, Zhu J, Liang J Wang Y-P, Wang H, Mao M Birth defects surveillance in China World J Pediatr 2Oll;7(4):3O2-31O doi:10.1007/sl2519-011-0326-0 52 Update on Overall Prevalence of Major 8irth Defects — Atlanta, Georgia, 1978 -2005 Accessed May 24,2021 https://www.cdc.gov/mmwr/previcw/mmwrhtml/mm5701a2.htm 53 Niculescu-Mizil A, Caruana R Predicting good probabilities with supervised learning In: Proceedings of the 22nd International Conference on Machine Learning ICML '05 Association for Computing Machinery; 2005:625-632 doi:10.1145/1102351.1102430 54 Wang L Discovering phase transitions with unsupervised learning Phys Rev B 2016;94( 19) 195105 dot: 10.1103/PhysRevB.94.195105 55 Noble ws What is a support vector machine? Nature Biotechnology 2006;24(12):1565- 1567 doi: 10.1038/nbt 1206-1565 56 James w Stein c Estimation with Quadratic Loss In: Kotz s, Johnson NL, eds Breakthroughs in Statistics: Foundations and Bosic Theory Springer Series in Statistics Springer; 1992:443- 460 doi:10.1007/978-l-4612-0919-5_30 57 Suykens JAK, Vandewalle J Least Squares Support Vector Machine Classifiers Neurol Processing Letters 1999;9(3):293300 doi: 10.102 3/A: 1018628609742 58 Grandjean H, Larroque D Levi S The performance of routine ultrasonographic screening of pregnancies in the Eurofetus Study American Journal of Obstetrics and Gynecology 1999;181(2):446 4S4 doi:10.1016/S0002-9378(99)70577 59 Sotiriadis A, Papatheodorou s, Eleftheriades M, Makrydimas G Nuchal translucency and major congenital heart defects in fetuses with normal karyotype: a meta-analysis Ultrasound in Obstetrics & Gynecology 2O13;42(4):383-389 dot :https://doi org/10.1002/uog 12488 60 Papatheodorou SI, Evangelou E, Makrydimas G, loannidis JPA first-trimester ductus venosus 4.655 screening for cardiac defects: a meta-anafysis BJOG: An International Journal of Obstetrics & Gynaecology 2011;118(12):1438-1445 4.656 doi:http$.//doi.org/10.1111/j.l471- 0528.201 l’o3O29.x 61 Rink BD, Norton ME Screening for fetal aneuploidy Seminars in Perinatology 2016;40(l):35-43 doi:10.1053/).semperi.201S 11.006 62 Wald NJ, Kennard A, Hackshaw A, McGuire A Antenatal Screening for Down's Syndrome J Med Screen 1997;4(4): 181-246 doi:10.1177/096914139700400402 63 Lao MR, Calhoun BC, Bracero LA, et al The Ability of the Quadruple Test to Predict Adverse Perinatal Outcomes in a High-risk Obstetric Population J Med Screen 2009;16(2):55-59 doi: 10.1258/jms.2009.009017 64 Cuckle H, Benn p, Pergamcnt E Maternal CÍDNA screening for Down syndrome-a cost sensitivity analysis Prenot Dtagn 2013;33(7):636-642 doi:10.1002/pd.4157 65 Spencer K, Ong CYT, Liao AWJ, Nicolaides KH The influence of ethnic origin on first trimester biochemical markers of chromosomal abnormalities Prenatal 0223(200006)20:63.0.CO;2-3 Diagnosis 2000;20(6):491- 494 doi:https://doi.org/10.1002/1097- 4.657 APPENDIX 4.658 Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISM A-ScR) Checklist 4.659 4.6604.661 4.663 SECTION 4.664 4.665 PRISMA-ScR CHECKLIST ITEM ITEM 4.667 TITLE 4.669 4.670 Identify the report as a scoping renew 4.668 Title 4.672 ABSTRACT 4.673 Stru ctured 4.674 sum 4.676 4.675 X nur 4.678 4.662 REPO RTED ON PAGE 4.671 Click here to enter text Provide a structured s unman* that includes (as applicable): badcgrotnd objectives, eligibility criteria, sources of evidence charting methods, results, and contusions that relate to the renew questions and objectives 4.677 Click heir to enter text INTRODUCTION 4.679 Rati onale 4.680 4.681 4.685 4.683 Obj ectives 4.684 4.687 METHODS 4.688 Pr otocol 4.690 ar*j 4.693 Eligi bility criteria 4.697 Infc Describe the raticnale f