In Vitro Fertilization (IVF) Cumulative Pregnancy Rate Prediction from Basic Patient Characteristics
Bo Zhang, Yuqi Cui, Meng Wang, Jingjing Li, Lei Jin, Dongrui Wu
aa r X i v : . [ q - b i o . Q M ] N ov In Vitro Fertilization (IVF) Cumulative PregnancyRate Prediction from Basic Patient Characteristics
Bo Zhang, Yuqi Cui, Meng Wang, Jingjing Li, Lei Jin and Dongrui Wu,
Abstract —Tens of millions of women suffer from infertilityworldwide each year. In vitro fertilization (IVF) is the bestchoice for many such patients. However, IVF is expensive, time-consuming, and both physically and emotionally demanding. Thefirst question that a patient usually asks before the IVF is howlikely she will conceive, given her basic medical examinationinformation. This paper proposes three approaches to predict thecumulative pregnancy rate after multiple oocyte pickup cycles.Experiments on 11,190 patients showed that first clustering thepatients into different groups and then building a support vectormachine model for each group can achieve the best overall perfor-mance. Our model could be a quick and economic approach forreliably estimating the cumulative pregnancy rate for a patient,given only her basic medical examination information, well beforestarting the actual IVF procedure. The predictions can help thepatient make optimal decisions on whether to use her own oocyteor donor oocyte, how many oocyte pickup cycles she may need,whether to use embryo frozen, etc. They will also reduce thepatient’s cost and time to pregnancy, and improve her quality oflife.
Index Terms —In vitro fertilization (IVF), machine learning,cumulative pregnancy rate prediction
I. I
NTRODUCTION
According to the World Health Organization (WHO) [33],infertility is “ a disease of the reproductive system defined bythe failure to achieve a clinical pregnancy after 12 months ormore of regular unprotected sexual intercourse. ” For womenunder 60, infertility was ranked the 5th highest serious globaldisability [1]. Estimates from 25 international population sur-veys sampling 172,413 women indicated that 9% of themsuffered from infertility [5]. Another study [14] on householdsurvey data from 277 demographic and reproductive healthsurveys for women aged 20-44 estimated that 48.5 millioncouples worldwide suffered from infertility in 2010. The 2006-2010 United States National Survey of Family Growth (NSFG)[7] sampling 22,682 men and women aged 15-44 also foundthat 6.0% (1.5 million) women suffered from infertility in2006-2010.Assisted reproductive technology (ART) [23] could helpthese couples to conceive pregnancy. The most common ART
B. Zhang, M. Wang, J. Li and L. Jin are with the ReproductiveMedicine Center, Tongji Hospital, Tongji Medical College, Huazhong Uni-versity of Science and Technology, Wuhan, Hubei 430030, China. Email:[email protected], tjmu mu [email protected], [email protected], [email protected]. Cui and D. Wu are with the Key Laboratory of the Ministry of Educationfor Image Processing and Intelligent Control, School of Artificial Intelligenceand Automation, Huazhong University of Science and Technology, Wuhan,Hubei 430074, China. Email: [email protected], [email protected] first two authors contributed equally to this work.Corresponding authors: Lei Jin ([email protected]), Dongrui Wu([email protected]). is in vitro fertilization (IVF) [8], which retrieves eggs froma woman’s ovaries, fertilizes them in the laboratory, andthen transfers the resulting embryos into the woman’s uterusthrough the cervix. According to the 2015 ART NationalSummary Report [2], more than 99% ART cycles performedin the United States in 2015 used IVF.The timeline of a typical IVF procedure is shown in Fig. 1.During the patient’s first visit, initial consultation is conducted,her medical history is recorded, and basic medical examinationis performed. This process may take 1-2 days. At Day 3,the patient’s basic characteristics such as age, BMI, infertilityduration, AFC, AMH, FSH, pathogenesis, etc., are available. Ifthe patient determines to perform IVF, then usually it will takethree menstrual cycles. In the first menstrual cycle, additionalexamination and controlled ovarian hyper-stimulation (COH)are performed. Oocyte pickup and egg fertilization are donein the second menstrual cycle. Embryo or balstocyst transferare performed in the third menstrual cycle. The entire processtakes about 2-3 months. During this process, embryo mor-phology features can be extracted to determine the embryoquality, number of embryo to transfer, and the transfer plan,etc. If the patient fails to conceive after embryo transfer, shehas to spend the same amount of time again to repeat thisprocedure, which represents a heavy burden to many patients,economically, physically, and emotionally.Cumulative pregnancy rate, which tells the probability thata patient conceives pregnancy after multiple IVF cycles, is animportant measure for evaluating different IVF approaches,and is usually also the first question that a patient asks beforestarting the IVF. Given the long duration (2-3 months) andhigh cost of an IVF cycle (the average cost of an IVF cycleis approximately $10,000-15,000 in the United States [12],and $4,500 in Tongji Hospital in China), it is important to beable to accurately estimate the individualized cumulative preg-nancy rate, so that the patient can make the most appropriatedecisions on whether to use her own oocyte or donor oocyte,how many oocyte pickup cycles she may need, whether to useembryo frozen, etc. Artificial intelligent, particularly machinelearning [4], could be used for this purpose.Machine learning has rapidly progressed the medical fieldduring the past few years. It has been used to predict thedevelopment of hepatocellular carcinoma [21], adult autismspectrum disorder [30], non-small cell lung cancer prognosis[32], human oocyte developmental potential [31], the risk ofacute myeloid leukaemia [3], etc., and also to identify a humanneonatal immune-metabolic network associated with bacterialinfection [22], to classify skin cancer [9], to isolate individualcell for scalable molecular genetic analysis of single cells [6],
Fig. 1. The IVF timeline. Our model utilizes only the basic medical examination information during the first visit, and it can give the cumulative pregnancyrate prediction on Day 3 when the initial medical examination results are ready. Conventional approaches in the literature use information during the actualIVF to predict the pregnancy rate, and hence are much more time-consuming and expensive than our approach. and so on.Machine learning has also been used to predict the preg-nancy result with features obtained before and during the IVF,including basic patient characteristics, embryo morphology,and so on. For example, decision trees [18], [19] have beenused to investigate the relationship between the outcome oftransfer and 53 embryo, oocyte and follicular features [20],to predict the IVF outcome from 100 variables related tothe basic patient characteristics (e.g., age, body mass index,etc.) and derived from the different stages of the IVF cycle(e.g., the amount of hormone treatment, the measurement ofovary volume, etc.) [17], and to predict the IVF outcome from69 features on patient’s basic information, diagnosis, clinicaltests, treatment methods, etc [11]. Bayesian classifiers havebeen used to select the most promising embryos to transferto the woman’s uterus using features related to clinical dataand embryo morphology [16], and to predict implantationoutcome of individual embryos in an IVF cycle from 18features including age, infertility factor, treatment protocol,sperm, embryo morphology, etc [25]. Support vector machines(SVMs) [27] and Bayesian Classifiers [26] have been usedto predict implantation outcomes of new embryos from 17features related to patient characteristics, clinical diagnosistreatment method, and embryo morphological parameters.However, to our knowledge, no one has used only patientcharacteristics from basic medical examinations to predict thecumulative IVF pregnancy rate, as we are doing in this study.In this paper, we propose supervised and unsupervisedmachine learning approaches for cumulative pregnancy rateprediction from basic patient characteristics. We show that theapproach that integrates unsupervised learning and supervisedlearning achieves the best performance. Our approach can sig-nificantly save the time and cost in predicting the cumulativeIVF pregnancy rate, and thus can help the patients make more appropriate decisions before the IVF starts.The remainder of this paper is organized as follows: Sec-tion II introduces our three machine learning approaches forcumulative pregnancy rate prediction. Section III presents theexperimental results. Section IV discusses the benefits of ourproposed approaches. Finally, Section V draws conclusion.II. O UR P ROPOSED M ACHINE L EARNING A PPROACHES
This section introduces the dataset used in our study,and the feature selection and machine learning approachesfor cumulative pregnancy rate prediction from basic patientcharacteristics.
A. The Dataset
This study consisted of 11,190 Chinese couples who suf-fered from infertility and received IVF treatments at TongjiHospital (ranked 3rd in Gynaecology and Obstetrics in China),Huazhong University of Science and Technology, Wuhan,China, between January 2016 and March 2018. Their IVFcycles varied from one to 11, as summarized in Table I. Onlybasic patient characteristics obtained from the initial medicalexamination were used in our prediction, which included fe-male age, female body mass index (BMI), infertility duration,antral follicle count (AFC), anti-mullerian hormone (AMH),follicle-stimulating hormone (FSH), and 30 pathogeny factors.
B. Feature Selection
In order to select the most informative features, we per-formed logistic regression [13] using all basic patient char-acteristics, where each categorical feature was converted to abinary value using one-hot encoding. We used only Cycle 1pregnancy results as the labels for logistic regression, andexcluded patients who did not receive a transfer in Cycle 1.
TABLE IS
UMMARY OF BASIC PATIENT CHARACTERISTICS IN OUR STUDY . T
HEFIRST SIX FEATURES ARE NUMERICAL . T
HEIR MEANS AND STANDARDDEVIATIONS ARE CALCULATED . P
ATHOGENY HAS FACTORS . F
OR EACHFACTOR , THE NUMBER OF PATIENTS AND THE PERCENTAGE ARE GIVEN .T HE USED FEATURES ARE MARKED BY ASTERISKS . Cycle Statistics
Cycle 1 n=9,419Cycle 2 n=1,432Cycle 3 n=236Cycle 4 n=59Cycle 5 n=30Cycle 6 n=7Cycle 7 n=2Cycle 8 n=2Cycle 9 n=2Cycle 10 n=0Cycle 11 n=1Total n=11,190
Patient Characteristics
Female age (years)* 31.5 ± kg/m )* 21.85 ± ± ± ng/ml )* 4.95 ± IU/L )* 7.94 ± Multiple logistic regression analyses showed that 14 featureshad significant correlation with pregnancy results (
P < . ).Among them, three etiological factors (endometrial tubercu-losis, chromosome abnormality, and others) had fewer than2% of the total patients. They were removed to make thefeatures more representative. As a result, 11 features werefinally selected for further analysis, and they are marked byasterisks in Table I. C. Cumulative Pregnancy Rate Prediction
The prediction of IVF outcome is extremely difficult usingonly basic patient characteristics without controlled ovar- ian hyper-stimulation details, and embryo and endometrialfeatures. According to previous research, embryo featuresare very important for the final outcome prediction usingmachine learning [11], [15]. When using only basic patientcharacteristics, we assume that patients having similar basiccharacteristics also have similar pregnancy rates. This is thebest assumption we could make before starting the actualIVF. When the patients start the IVF, more features could beextracted, and more individualized prediction could be made.However, these features are not available before the IVF, andhence will not be used in our model.We constructed three different machine learning models –clustering, SVM, and clustering-SVM (C-SVM), and com-pared their performances using three measures. The pipelineof our three machine learning approaches is shown in Fig. 2.Only the 11 asterisk features in Table I were used. We firstused one-hot encoding to convert each categorical feature intonumerical features, and then performed z -normalization totransform each feature to have mean 0 and standard deviation1. D. Model 1: Clustering
In the training phase of the clustering approach, we firstapplied k -means clustering with k = 30 to all patients. Wethen identified all possible × / unique pairs ofclusters. For each pair, we performed the log-rank test [10],[24], [29] between the two clusters to check if the differencebetween them was significant. If the p value of at least one ofthe 435 tests was larger than a predefined threshold α ( α =0 . was used in our study), then we identified the two clusterswith the largest p -value (which meant the two clusters were themost similar) and merged them. We repeated the log-rank testswith the remaining clusters, until all p -values were smallerthan α . We then recorded the center of each cluster, and itscorresponding cumulative pregnancy rate.In the testing phase, when the basic characteristics of a newpatient came in, we assigned the patient to the cluster with theclosest centroid, and then used the corresponding cumulativepregnancy rate as the prediction. E. Model 2: SVM
For the SVM classifier [28], we first performed 5-foldcross validation on the training set to search for the bestkernel function (polynomial, RBF, or linear) and to determinewhether a larger weight should be used to accommodate theminority class. Eventually we used the RBF kernel and set theper-class weights inversely proportional to class frequencies inthe training data. We then used penalty parameter C = 1 totrain a probabilistic SVM classifier. F. Model 3: C-SVM
The C-SVM approach was a sequential combination of theclustering approach and the SVM approach. In the trainingphase, it first used the clustering approach to group the patientsinto several clusters, and then trained an RBF SVM for eachcluster to individualize the patients within each cluster.
Fig. 2. Pipeline of the three proposed machine learning approaches. Given a training dataset of patients with basic medical examination information, all threemodels use the 11 asterisk features in Table I, convert the categorical features to numerical features using one-hot encoding, and z -normalize each feature.Clustering is an unsupervised approach. SVM is a supervised approach. C-SVM integrates both unsupervised and supervised approaches. In the testing phase, when the basic characteristics of a newpatient came in, we first assigned the patient to the cluster withthe closest centroid, and then used the corresponding SVM topredict a more individualized cumulative pregnancy rate.III. P
REDICTION R ESULTS
This section compares the prediction performances of thethree proposed approaches.
A. Area under the Curve (AUC)
First, we evaluated the performances of the three approachesby randomly sampling two thirds of the patients as trainingdata, and the remaining one third as test data. We used thetraining data to train the three models and then validated themon the test data. Their receiver operating characteristic (ROC)curves are shown in Fig. 3, and the corresponding areas underthe curve (AUCs) were also computed and indicated in thelegend. Fig. 3 shows that SVM and C-SVM had similar AUCperformances (0.69 and 0.70, respectively), both of which werehigher than clustering (AUC=0.67).
B. Cumulative Pregnancy Rate Prediction
Once we get the predicted probability and the correspond-ing cluster of each patient in test data, we can predict thecumulative pregnancy rate using the mean probability of thecorresponding cluster. Fig. 4 shows the cumulative pregnancyrate curve using the three approaches. Although SVM hadpromising AUC in Fig. 3, its cumulative pregnancy rateprediction had large biases. On the other hand, clustering andC-SVM, particularly C-SVM, had much smaller predictionerrors. T r u e p o s i t i v e r a t e ROC curves clustering (AUC = 0.67)SVM (AUC = 0.69)C-SVM (AUC = 0.70)
Fig. 3. ROC curves and AUCs of the three approaches.
C. Stability of the Prediction Models
In order to test the stability of the three prediction models,we repeated them 30 times, each time with different trainingand test data. As Table I shows that less than 1% patients hadmore than three cycles, we did not consider cycle numberslarger than three. The mean and standard deviation of theAUCs from the 30 runs are shown in the first part of Table II.On average C-SVM achieved the best AUCs in the threecycles.We also studied the stability of the three approaches us-ing another the root mean squared error (RMSE). For eachmodel in each run, we concatenated the predicted cumulativepregnancy rates in three cycles and n clusters into a n -element vector ˆ y = [ˆ y , ..., ˆ y n ] , and computed the RMSE P r o b a b ili t y ( % ) Clustering (a) P r o b a b ili t y ( % ) SVM (b) P r o b a b ili t y ( % ) C-SVM (c)Fig. 4. Predicted cumulative pregnancy rate curves of the three models ontest data. (a) Clustering; (b) SVM; (c) C-SVM. The solid curves in the samecolor in the three subfigures are identical, indicating the true cumulativepregnancy rate curve of a cluster. The dashed curves indicate the predictionsfrom different approaches. In each subfigure, the solid and dashed curves inthe same color indicate results in the same cluster, which should be close. TABLE IIM
EAN AND STANDARD DEVIATION ( IN PARENTHESES ) OF AUC
S AND
RMSE
S OF THE THREE PROPOSED APPROACHES . T
HE BEST ONES AREMARKED IN BOLD .Cycle Clustering SVM C-SVMAUC 1 0.6725 (0.0075) 0.6906 (0.0080)
RMSE All 0.0274 (0.0098) 0.0794 (0.0130) between the predictions and the corresponding groundtruth y = [ y , ..., y n ] , RM SE = " n n X i =1 (ˆ y i − y i ) / (1)A smaller RMSE means a better performance. The mean andstd of the RMSEs in the 30 runs are shown in the second partof Table II. Again, C-SVM achieved the best performance.We also performed an analysis of variance (ANOVA) test tocheck if there was statistically significant difference betweeneach pair of algorithms. The p -values are shown in Table III,where the statistically significant ones are marked in bold.SVM and C-SVM were statistically significantly better thanclustering on AUC, and SVM and C-SVM were statisticallysignificantly better than clustering on RMSE. In summary, C-SVM achieved the best overall performance. TABLE III p - VALUES OF
ANOVA
TESTS ( α = 0 . ) ON THE THREE PROPOSEDAPPROACHES .Clustering SVMAUC SVM
C-SVM
C-SVM 0.97
IV. D
ISCUSSIONS
This section discusses the advantages of our proposedapproaches, particulary C-SVM, our best-performing model.
A. C-SVM Reduce the Time and Cost to Predict the IVFCumulative Pregnancy Rate
Our C-SVM model uses only the basic medical examinationinformation during the first visit (which takes two days andcosts about $50 in Tongji Hospital in China) to predict thecumulative pregnancy rate, and the result can be knownimmediately after the visit.Compared with the conventional approaches in the liter-ature, which use information during the IVF (which takes2-3 months and costs about $4,500 in Tongji Hospital inChina), our approach is much faster and more economic. Itsignificantly saves the patient’s time and cost, and represents astep towards precision medicine and individualized treatment.
B. Cumulative Pregnancy Rate Prediction Is More Informativethan Single-Cycle Pregnancy Rate Prediction
The total duration and cost of IVF is significantly impactedby the number of oocyte pickup cycles. Since oocyte pickupis time-consuming and expensive, a patient may choose tofreeze the extra embryos from the first oocyte pickup cycle toreduce the time and cost: the frozen embryos can be transferredin case previous transfers fail, without the need to pickupfresh oocyte and fertilize them again. However, frozen embryotransfer may have a lower pregnancy rate than fresh embryotransfer. So, it is important to know the cumulative pregnancyrate of fresh embryo transfers so that the patient can makea smarter decision on whether it is worthwhile to save thetime and cost of another oocyte pickup cycle. Our C-SVMmodel can predict the cumulative pregnancy rates after one,two or three oocyte pickup cycles, which gives the patientsexact information they need in decision making.A patient with poor ovarian reserve is very difficult toconceive using her own oocyte. Knowing the cumulativepregnancy rate using her own oocyte could greatly help hermake a wiser decision: if the cumulative pregnancy rateusing her own oocyte is much lower than her expectation,then the patient may choose to receive donor oocyte, whichmay have much higher pregnancy rate. In this way, thepatient can avoid potentially multiple controlled ovarian hyper-stimulations, shorten the time to pregnancy, reduce the overallcost, and hence improve the quality of life.V. C
ONCLUSION
In this paper, we have developed three different approaches(clustering, SVM, and C-SVM) to predict the IVF cumulativepregnancy rate in multiple cycles of oocyte pickup usingbasic patient characteristic. The selected parameters includedfemale age, female BMI, infertility duration, AFC, AMH,FSH, and five pathogeny factors (diminished ovarian reserve,perimenopause, paternal factor, PCOS, and intrauterine adhe-sion). Experimental results showed that the AUCs of SVM andC-SVM were better than that of clustering, and the predictionRMSEs of clustering and C-SVM were smaller than that ofSVM. In summary, C-SVM seems to be the best model.To our best knowledge, this is the first study on usingmachine learning to predict the cumulative pregnancy rate ofmultiple IVF cycles from only basic patient characteristicsbefore the actual IVF. The predictions can help the patientmake optimal decisions on whether to use her own oocyte ordonor oocyte, how many oocyte pickup cycles she may need,whether to use embryo frozen, etc. They will also reduce thepatient’s cost and time to pregnancy, and improve her qualityof life. R et al. , “Prediction of acute myeloid leukaemia risk in healthy individu-als,”
Nature , p. 1, 2018. [4] C. M. Bishop,
Pattern Recognition and Machine Learning . NY:Springer-Verlag, 2006.[5] J. Boivin, L. Bunting, J. A. Collins, and K. G. Nygren, “Internationalestimates of infertility prevalence and treatment-seeking: potential needand demand for infertility medical care,”
Human Reproduction , vol. 22,no. 6, pp. 1506–1512, 2007.[6] C. Brasko, K. Smith, C. Molnar, N. Farago, L. Hegedus, A. Balind,T. Balassa, A. Szkalisity, F. Sukosd, K. Kocsis et al. , “Intelligent image-based in situ single-cell isolation,”
Nature Communications , vol. 9, no. 1,p. 226, 2018.[7] A. Chandra, C. Copen, and E. Stephen, “Infertility andimpaired fecundity in the united states, 1982-2010: Datafrom the national survey of family growth,”
National HealthStatistics Reports
Reproductive Biologyand Endocrinology , vol. 16, no. 1, p. 64, 2018.[9] A. Esteva, B. Kuprel, R. A. Novoa, J. Ko, S. M. Swetter, H. M. Blau,and S. Thrun, “Dermatologist-level classification of skin cancer withdeep neural networks,”
Nature , vol. 542, no. 7639, p. 115, 2017.[10] M. K. Goel, P. Khanna, and J. Kishore, “Understanding survival analysis:Kaplan-Meier estimate,”
International journal of Ayurveda research ,vol. 1, no. 4, p. 274, 2010.[11] R.-S. Guh, T.-C. J. Wu, and S.-P. Weng, “Integrating genetic algorithmand decision tree learning for assistance in predicting in vitro fertilizationoutcomes,”
Expert Systems with Applications , vol. 38, no. 4, pp. 4437–4449, 2011.[12] M. D. Hornstein, “State of the ART: assisted reproductive technologiesin the United States,”
Reproductive Sciences , vol. 23, no. 12, pp. 1630–1633, 2016.[13] D. W. Hosmer Jr, S. Lemeshow, and R. X. Sturdivant,
Applied logisticregression . John Wiley & Sons, 2013, vol. 398.[14] M. N. Mascarenhas, S. R. Flaxman, T. Boerma, S. Vanderpoel, andG. A. Stevens, “National, regional, and global trends in infertilityprevalence since 1990: A systematic analysis of 277 health surveys,”
PLOS Medicine , vol. 9, no. 12, pp. 1–12, 12 2012.[15] R. Milewski, P. Malinowski, A. J. Milewska, J. Czerniecki, P. Ziniewicz,and S. Wołczy´nski, “Nearest neighbor concept in the study of IVFICSI/ET treatment effectiveness,”
Studies in Logic, Grammar andRhetoric. Logical, Statistical and Computer Methods in Medicine ,vol. 25, no. 38, pp. 49–57, 2011.[16] D. A. Morales, E. Bengoetxea, P. Larra˜naga, M. Garc´ıa, Y. Franco,M. Fresnada, and M. Merino, “Bayesian classification for the selectionof in vitro human embryos using morphological and clinical data,”
Computer Methods and Programs in Biomedicine , vol. 90, no. 2, pp.104–116, 2008.[17] L. Passmore, J. Goodside, L. Hamel, L. Gonzalez, T. Silberstein,and J. Trimarchi, “Assessing decision tree models for clinical in-vitrofertilization data,” Dept. of Computer Science and Statistics, Universityof Rhode Island, TR03-296, Tech. Rep., 2003.[18] J. R. Quinlan, “Induction of decision trees,”
Machine Learning , vol. 1,no. 1, pp. 81–106, 1986.[19] J. R. Quinlan,
C4.5: programs for machine learning . San Francisco,CA: Morgan Kaufmann Publishers Inc., 1992.[20] R. Saith, A. Srinivasan, D. Michie, and I. L. Sargent, “Relationshipsbetween the developmental potential of human in-vitro fertilizationembryos and features describing the embryo, oocyte and follicle,”
Human Reproduction Update , vol. 4, no. 2, pp. 121–134, 1998.[21] A. G. Singal, A. Mukherjee, B. J. Elmunzer, P. D. Higgins, A. S. Lok,J. Zhu, J. A. Marrero, and A. K. Waljee, “Machine learning algorithmsoutperform conventional regression models in predicting development ofhepatocellular carcinoma,”
The American Journal of Gastroenterology ,vol. 108, no. 11, p. 1723, 2013.[22] C. L. Smith, P. Dickinson, T. Forster, M. Craigon, A. Ross, M. R. Khon-doker, R. France, A. Ivens, D. J. Lynn, J. Orme et al. , “Identification ofa human neonatal immune-metabolic network associated with bacterialinfection,”
Nature Communications , vol. 5, p. 4649, 2014.[23] S. Sunderam, D. M. Kissin, S. B. Crawford, S. G. Folger, D. J. Jamieson,L. Warner, and W. D. Barfield, “Assisted reproductive technologysurveillance – United States, 2014,”
MMWR Surveillance Summaries ,vol. 66, no. 6, pp. 1–24, 2017.[24] S. L. Tan, J. Betts, B. Mason, R. Edwards, S. Campbell, P. Royston,and H. Jacobs, “Cumulative conception and livebirth rates after In-VitroFertilisation,”
The Lancet , vol. 339, no. 8806, pp. 1390–1394, 1992. [25] A. Uyar, A. Bener, and H. N. Ciray, “Predictive modeling of implantationoutcome in an in vitro fertilization setting: an application of machinelearning methods,”
Medical Decision Making , vol. 35, no. 6, pp. 714–725, 2015.[26] A. Uyar, A. Bener, H. N. Ciray, and M. Bahceci, “ROC based evaluationand comparison of classifiers for IVF implantation prediction,” in
International Conference on Electronic Healthcare . Istanbul, Turkey:Springer, Sep. 2009, pp. 108–111.[27] A. Uyar, H. N. Ciray, A. Bener, and M. Bahceci, “3P: Personalizedpregnancy prediction in IVF treatment process,” in
International Con-ference on Electronic Healthcare . London, UK: Springer, Sep. 2008,pp. 58–65.[28] V. Vapnik,
Statistical Learning Theory . New York, NY: Wiley Press,1998.[29] V. J. Verwaal, S. van Ruth, E. de Bree, G. W. van Slooten, H. vanTinteren, H. Boot, and F. A. Zoetmulder, “Randomized trial of cytore-duction and hyperthermic intraperitoneal chemotherapy versus systemicchemotherapy and palliative surgery in patients with peritoneal carci-nomatosis of colorectal cancer,”
Journal of clinical oncology , vol. 21,no. 20, pp. 3737–3743, 2003.[30] N. Yahata, J. Morimoto, R. Hashimoto, G. Lisi, K. Shibata,Y. Kawakubo, H. Kuwabara, M. Kuroda, T. Yamada, F. Megumi et al. ,“A small number of abnormal brain connections predicts adult autismspectrum disorder,”
Nature Communications , vol. 7, p. 11254, 2016.[31] L. Z. Yanez, J. Han, B. B. Behr, R. A. R. Pera, and D. B. Camarillo,“Human oocyte developmental potential is predicted by mechanicalproperties within hours after fertilization,”
Nature Communications ,vol. 7, p. 10809, 2016.[32] K.-H. Yu, C. Zhang, G. J. Berry, R. B. Altman, C. R´e, D. L. Rubin, andM. Snyder, “Predicting non-small cell lung cancer prognosis by fullyautomated microscopic pathology image features,”
Nature Communica-tions , vol. 7, p. 12474, 2016.[33] F. Zegers-Hochschild, G. D. Adamson, J. de Mouzon, O. Ishihara,R. Mansour, K. Nygren, E. Sullivan, and S. Vanderpoel, “Internationalcommittee for monitoring assisted reproductive technology (ICMART)and the world health organization (WHO) revised glossary of ARTterminology, 2009,”