Embryo selection through artificial intelligence versus embryologists: a systematic review

Human Reproduction Open, Journal Year: 2023, Volume and Issue: 2023(3)

Published: Jan. 1, 2023

What is the present performance of artificial intelligence (AI) decision support during embryo selection compared to standard by embryologists?AI consistently outperformed clinical teams in all studies focused on morphology and outcome prediction assessment.The ART success rate ∼30%, with a worrying trend increasing female age correlating considerably worse results. As such, there have been ongoing efforts address this low through development new technologies. With advent AI, potential for machine learning be applied such manner that areas limited human subjectivity, as selection, can enhanced increased objectivity. Given AI improve IVF rates, it remains crucial review between embryologists selection.The search was done across PubMed, EMBASE, Ovid Medline, IEEE Xplore from 1 June 2005 up including 7 January 2022. Included articles were also restricted those written English. Search terms utilized databases study were: ('Artificial intelligence' OR 'Machine Learning' 'Deep learning' 'Neural network') AND ('IVF' 'in vitro fertili*' 'assisted reproductive techn*' 'embryo'), where character '*' refers engine include any auto completion term.A literature conducted relating applications IVF. Primary outcomes interest accuracy, sensitivity, specificity grade assessments likelihood outcomes, pregnancy after treatments. Risk bias assessed using Modified Down Black Checklist.Twenty included review. There no specific assessment day studies-Day until Day 5/6 investigated. The types input training algorithms images time-lapse (10/20), information (6/20), both (4/20). Each model demonstrated promise when an embryologist's visual assessment. On average, models predicted successful greater accuracy than embryologists, signifying reliability prediction. performed at median 75.5% (range 59-94%) predicting grade. correct (Ground Truth) defined use according post embryologists' following local respective guidelines. Using blind test datasets, 65.4% 47-75%) same ground truth provided original Similarly, had 77.8% 68-90%) patient treatment 64% 58-76%) embryologists. When images/time-lapse inputs combined, higher 81.5% 67-98%), while 51% 43-59%).The findings are based not prospectively evaluated setting. Additionally, fair comparison deemed unfeasible owing heterogeneity studies, models, database employed design quality.AI provides considerable field selection. However, needs shift developers' perception implantation towards or live birth. existing focus locally generated many lack external validation.This funded Monash Data Future Institute. All authors conflicts declare.CRD42021256333.

Language: Английский

---

An artificial intelligence model correlated with morphological and genetic features of blastocyst quality improves ranking of viable embryos

Reproductive BioMedicine Online, Journal Year: 2022, Volume and Issue: 45(6), P. 1105 - 1117

Published: Aug. 3, 2022

Language: Английский

---

Development and validation of deep learning based embryo selection across multiple days of transfer

Scientific Reports, Journal Year: 2023, Volume and Issue: 13(1)

Published: March 14, 2023

Abstract This work describes the development and validation of a fully automated deep learning model, iDAScore v2.0, for evaluation human embryos incubated 2, 3, 5 or more days. We trained evaluated model on an extensive diverse dataset including 181,428 from 22 IVF clinics across world. To discriminate transferred with known outcome, we show areas under receiver operating curve ranging 0.621 to 0.707 depending day transfer. Predictive performance increased over time showed strong correlation morphokinetic parameters. The model’s is equivalent KIDScore D3 3 while it significantly surpasses D5 v3 5+ embryos. provides analysis time-lapse sequences without need user input, reliable method ranking their likelihood implantation, at both cleavage blastocyst stages. greatly improves embryo grading consistency saves compared traditional methods.

Language: Английский

---

Associations between the artificial intelligence scoring system and live birth outcomes in preimplantation genetic testing for aneuploidy cycles

Reproductive Biology and Endocrinology, Journal Year: 2024, Volume and Issue: 22(1)

Published: Jan. 17, 2024

Abstract Background Several studies have demonstrated that iDAScore is more accurate in predicting pregnancy outcomes cycles without preimplantation genetic testing for aneuploidy (PGT-A) compared to KIDScore and the Gardner criteria. However, effectiveness of with PGT-A has not been thoroughly investigated. Therefore, this study aims assess association between artificial intelligence (AI)-based (version 1.0) single-embryo transfer (SET) PGT-A. Methods This retrospective was approved by Institutional Review Board Chung Sun Medical University, Taichung, Taiwan. Patients undergoing SET ( n = 482) following at a single reproductive center January 2017 June 2021. The blastocyst morphology morphokinetics all embryos were evaluated using time-lapse system. blastocysts ranked based on scores generated iDAScore, which defined as AI scores, or D5 3.2) manufacturer’s protocols. A transferred after examining embryonic ploidy status next-generation sequencing-based platform. Logistic regression analysis generalized estimating equations conducted whether are associated probability live birth (LB) while considering confounding factors. Results revealed score significantly LB (adjusted odds ratio [OR] 2.037, 95% confidence interval [CI]: 1.632–2.542) when pulsatility index (PI) level types chromosomal abnormalities controlled. Blastocysts divided into quartiles accordance their (group 1: 3.0–7.8; group 2: 7.9–8.6; 3: 8.7–8.9; 4: 9.0–9.5). Group 1 had lower rate (34.6% vs. 59.8–72.3%) higher loss (26% 4.7–8.9%) other groups p < 0.05). receiver operating characteristic curve verified significant but limited ability predict (area under [AUC] 0.64); weaker than combination type abnormalities, PI (AUC 0.67). In comparison non-LB groups, both euploid (median: 8.6 8.8) mosaic 8.0 8.6) SETs. Conclusions Although its predictive can be further enhanced, cycles. Euploid low (≤ 7.8) rate, indicating potential annotation-free system decision-support tool deselecting poor

Language: Английский

---

Embryologist agreement when assessing blastocyst implantation probability: is data-driven prediction the solution to embryo assessment subjectivity?

Human Reproduction, Journal Year: 2022, Volume and Issue: 37(10), P. 2275 - 2290

Published: Aug. 9, 2022

What is the accuracy and agreement of embryologists when assessing implantation probability blastocysts using time-lapse imaging (TLI), can it be improved with a data-driven algorithm?The overall interobserver large panel was moderate prediction modest, while purpose-built artificial intelligence model generally resulted in higher performance metrics.Previous studies have demonstrated significant variability amongst embryo quality. However, data concerning embryologists' ability to predict TLI still lacking. Emerging technologies based on tools shown great promise for improving selection predicting clinical outcomes.TLI video files 136 embryos known were retrospectively collected from two sites between 2018 2019 assessment 36 comparison deep neural network (DNN).We recruited 39 13 different countries. All participants blinded outcomes. A total videos that reached blastocyst stage used this experiment. Each embryo's likelihood successfully implanting assessed by embryologists, providing grades (IPGs) 1 5, where indicates very low 5 high likelihood. Subsequently, three over years experience provided Gardner scores. categorized into quality groups their Embryologist predictions then converted (IPG ≥ 3) no ≤ 2). Embryologists' Fleiss kappa coefficient. 10-fold cross-validation DNN developed provide IPGs files. The model's compared embryologists.Logistic regression employed following confounding variables: country residence, academic level, scoring system, log TLI. None found statistically impact embryologist at α = 0.05. average 51.9% all (N 136). top poor (according score categorizations) 57.5% 57.4%, respectively, 44.6% fair embryos. Overall (κ 0.56, N best achieved + group 0.65, 77), lower 0.25, 59). showed an rate 62.5%, accuracies 62.2%, 61% 65.6% poor, groups, respectively. AUC than (0.70 vs 0.61 embryologists) as well (DNN embryologists-Poor: 0.69 0.62; Fair: 0.67 0.53; Top: 0.77 0.54).Blastocyst performed acquired incubators, each contained single focal plane. Clinical regarding underlying cause infertility endometrial thickness before transfer not available, yet may explain failure IPGs. Implantation defined presence gestational sac, whereas detection fetal heartbeat more robust marker viability. raw anonymized extent possible quantify number unique patients cycles included study, potentially masking effect bias limited patient pool. Furthermore, lack demographic makes difficult draw conclusions how representative dataset wider population. Finally, required assess potential, Although traditional approach evaluation, morphology/morphokinetics means believed strongly correlated viability and, some methods, potential.Embryo key element IVF success continues challenge. Improving predictive could assist optimizing rates other outcomes minimize financial emotional burden patient. This study demonstrates likely due subjective nature assessment. In particular, we significantly Using algorithms assistive tool help professionals increase promote much needed standardization clinic. Our results indicate need further research technological advancement field.Embryonics Ltd Israel-based company. Funding partially Israeli Innovation Authority, grant #74556.N/A.

Language: Английский

---

Development and evaluation of a live birth prediction model for evaluating human blastocysts from a retrospective study

eLife, Journal Year: 2023, Volume and Issue: 12

Published: Feb. 22, 2023

In infertility treatment, blastocyst morphological grading is commonly used in clinical practice for evaluation and selection, but has shown limited predictive power on live birth outcomes of blastocysts. To improve prediction, a number artificial intelligence (AI) models have been established. Most existing AI only images the area under receiver operating characteristic (ROC) curve (AUC) achieved by these plateaued at ~0.65.This study proposed multimodal method using both patient couple's features (e.g., maternal age, hormone profiles, endometrium thickness, semen quality) to predict human utilize data, we developed new model consisting convolutional neural network (CNN) process multilayer perceptron features. The data set this consists 17,580 blastocysts with known outcomes, images, features.This an AUC 0.77 which significantly outperforms related works literature. Sixteen out 103 were identified be predictors helped prediction. Among features, day transfer, antral follicle count, retrieved oocyte number, thickness measured before transfer are top five contributing Heatmaps showed that CNN mainly focuses image regions inner cell mass trophectoderm (TE) contribution TE-related was greater trained inclusion compared alone.The results suggest along increases prediction accuracy.Natural Sciences Engineering Research Council Canada Chairs Program.More than 50 million couples worldwide experience infertility. most common treatment vitro fertilization (IVF). Fertility specialists collect eggs sperm from prospective parents. They combine egg laboratory allow fertilized develop days into multi-celled blastocyst. Then, select healthiest return them patient's uterus. Since 1978, more 8 children conceived through IVF. Yet, about 30% IVF attempts result successful birth. As result, fertility patients often undergo multiple rounds IVF, can expensive emotionally draining. Several factors determine success, one health selected Specialists several criteria. But assessments subjective inconsistent predicting ones likely Recent studies technology may help Liu et al. show assess characteristics leads accurate predictions experiments, researchers computer program pictures parents' characteristics. 16 parental associated outcomes. 5 uterus, how many present ovaries, uterus lining. highest healthy births so far, success rates listed other studies. Artificial intelligence-aided blastocyte selection reduce cycles undergo. Before use their clinics, they must conduct confirmatory enroll compare conventional methods intelligence.

Language: Английский

---

A brief history of artificial intelligence embryo selection: from black-box to glass-box

Human Reproduction, Journal Year: 2023, Volume and Issue: 39(2), P. 285 - 292

Published: Dec. 7, 2023

Abstract With the exponential growth of computing power and accumulation embryo image data in recent years, artificial intelligence (AI) is starting to be utilized selection IVF. Amongst different AI technologies, machine learning (ML) has potential reduce operator-related subjectivity while saving labor time on this task. However, as modern deep (DL) techniques, a subcategory ML, are increasingly used, its integrated black-box attracts growing concern owing well-recognized issues regarding lack interpretability. Currently, there randomized controlled trials confirm effectiveness such models. Recently, emerging evidence shown underperformance models compared more interpretable traditional ML selection. Meanwhile, glass-box AI, being promoted across wide range fields supported by ethical advantages technical feasibility. In review, we propose novel classification system for AI-driven systems from an embryology standpoint, defining morphology-based approaches with emphasis subjectivity, explainability,

Language: Английский

---

Improved prediction of clinical pregnancy using artificial intelligence with enhanced inner cell mass and trophectoderm images

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Feb. 8, 2024

Abstract This study aimed to assess the performance of an artificial intelligence (AI) model for predicting clinical pregnancy using enhanced inner cell mass (ICM) and trophectoderm (TE) images. In this retrospective study, we included static images 2555 day-5-blastocysts from seven in vitro fertilization centers South Korea. The main outcome was predictive capability detect pregnancies (gestational sac). Compared with original embryo images, use ICM TE improved average area under receiver operating characteristic curve AI 0.716 0.741. Additionally, a gradient-weighted class activation mapping analysis demonstrated that image-trained able extract features crucial areas 99% (506/512) cases. Particularly, it could TE. contrast, trained on focused only 86% (438/512) Our results highlight potential efficacy ICM- TE-enhanced when training models predict pregnancy.

Language: Английский

---

An artificial intelligence algorithm for automated blastocyst morphometric parameters demonstrates a positive association with implantation potential

Scientific Reports, Journal Year: 2023, Volume and Issue: 13(1)

Published: Sept. 5, 2023

Abstract Blastocyst selection is primarily based on morphological scoring systems and morphokinetic data. These methods involve subjective grading time-consuming techniques. Artificial intelligence allows for objective quick blastocyst selection. In this study, 608 blastocysts were selected transfer using morphokinetics Gardner criteria. Retrospectively, morphometric parameters of size, inner cell mass (ICM) ICM-to-blastocyst size ratio, ICM shape automatically measured by a semantic segmentation neural network model. The model was trained 1506 videos with 102 validation no overlap between the trophectoderm models. Univariable logistic analysis found ratio to be significantly associated implantation potential. Multivariable regression analysis, adjusted woman age, odds increased 1.74 embryos greater than mean (147 ± 19.1 μm). performance algorithm represented an area under curve 0.70 (p < 0.01). conclusion, study supports association large higher potential suggests that morphometrics can used as precise, consistent, time-saving tool improving

Language: Английский

---

Introducing artificial intelligence and sperm epigenetics in the fertility clinic: a novel foundation for diagnostics and prediction modelling

Frontiers in Reproductive Health, Journal Year: 2025, Volume and Issue: 7

Published: Feb. 27, 2025

Worldwide, infertility is a rising problem. A couple's lifestyle, age and environmental exposures can interfere with reproductive health. The scientific field tries to understand the various processes how male female factors may affect fertility, but translation clinic limited. I here emphasize potential reasons for failure in optimal treatment planning especially why current prediction modelling falls short. First, Assisted Reproductive Technology (ART) has become mainstream solution couples experiencing infertility, while causes of remain unexplored or undetermined. For instance, role men generally left out preconceptional testing care. Second, regularly used statistical computational methods estimate pregnancy outcomes miss important biological factors, including features from side (e.g., age, smoking, obesity status, alcohol use occupation), as well genetic epigenetic characteristics. suggest using an integrated approach biostatistics machine learning improve diagnostics fertility clinic. novelty this concept includes empirically collected information on sperm epigenome combined readily available data medical records both partners lifestyle factors. As needs well-designed models at different levels, derivatives are needed. objectives patients, clinicians, embryologists differ slightly, mathematical need be adapted accordingly. multidisciplinary where patients seen by both, clinicians biomedically skilled counsellors, could help provide evidence-based assistance success. Next, when it concerns that change ability produce embryos ART, embryologist would benefit personalized model, history patient easily accessible germ cells, such sperm.

Language: Английский

---