Sperm preparation techniques are essential for enhancing fertilization potential by isolating motile, viable sperm while removing harmful components such as debris, reactive oxygen species (ROS), and immotile sperm. These methods are fundamental to assisted reproductive technologies (ART), optimizing outcomes in procedures like intrauterine insemination (IUI) and in vitro fertilization (IVF). Traditionally, sperm processing has been categorized into three primary techniques: density gradient centrifugation, swim-up, and simple wash. This chapter provides detailed protocols for these classic methods while exploring advanced techniques such as microfluidics, which offer improved precision and efficiency in sperm selection for ART applications.

Artificial intelligence (AI) is generating genuine excitement in reproductive medicine for good reason. The promise of more consistent embryo assessment, reduced inter-observer variability, better outcome prediction and meaningful support for the embryologist's judgment represents a real opportunity to improve care for patients who have already invested enormously in the pursuit of a family. Realizing that promise, however, depends on something the field has not always done well which is insisting on rigorous validation before widespread adoption. This editorial is written in that spirit, as advocates for AI who believe the technology's long-term acceptance by patients, clinicians and embryologists will be determined by how carefully we deploy it now and how we proceed going forward. How will we know when AI is ready to fully deploy in the IVF lab? The answer requires distinguishing between being safe and being safe and effective and the medical community has not yet reached consensus on either standard for most applications. Professional societies will need to lead that process. However, in the meantime, with commercial adoption accelerating and stakeholder incentives not always aligned with patient welfare, we propose criteria as a starting framework for that discussion.

INTRODUCTION:Ovarian stimulation (OS) alters the peri-implantation environment and may affect placentation. Therefore, our study aimed to compare placental pathology and PW among singleton live births conceived after IUI with gonadotropins (Gn), oral medications (OM; clomiphene/letrozole), or unstimulated/natural (Nat) cycles. METHODS:Retrospective review of 386 IUI(±OS) singleton live births with placental examination: Gn (n = 222), OM (n = 129), and Nat (n = 35). Outcomes were placental lesions (classified as anatomic, inflammatory, infectious, or vascular) and PW. Generalized estimating equations (GEE) estimated adjusted risk ratios (adjRR) for pathology, adjusted beta (adjβ), for PW, and their respective 95% confidence interval (95%CI), controlling for maternal age, BMI, race, PCOS diagnosis, gestational age at delivery, infant sex, and pregnancy complications (hypertensive disorders and diabetes). Sensitivity analyses included uncomplicated term livebirths and inverse probability weighting (IPW). RESULTS:After multivariable GEE adjustment, OM was associated with higher odds of anatomic abnormalities compared with both Gn and Nat [adjRR (95% CI): 1.15 (1.02, 1.31) and 1.28 (1.05, 1.56), respectively], whereas Gn did not differ compared to Nat. In our sensitivity analyses, adjRR (95%CI) for anatomic lesions remained similarly more prevalent in OM compared to Nat [Uncomplicated: 1.21 (0.96, 1.52); IPW: 1.29 (1.02, 1.63)], while it was not significant compared to Gn [Uncomplicated: 1.09 (0.93, 1.29); IPW: 1.12 (0.97, 1.30)]. Inflammatory, infectious, vascular placental abnormalities, and PW remained comparable between groups in the main analysis and after the same restrictions. CONCLUSION:Among IUI-conceived singleton births with placental evaluation, oral medications were associated with more anatomic placental abnormalities than gonadotropins or natural cycles, whereas PW did not differ by protocol.

RESEARCH QUESTION:Does the combination of fast vitrification and fast warming protocols, compared with standard vitrification and standard warming protocols, result in differences in pregnancy outcomes following frozen embryo transfer (FET)? DESIGN:This was a retrospective comparative cohort study of unique patients undergoing either natural, medicated or programmed FET cycles of single blastocysts. The first FET following fresh cycles was included for each patient. Three groups were analysed: standard vitrification with standard warming protocol (n = 1025); standard vitrification with fast warming protocol (n = 926); and fast vitrification with fast warming protocol (n = 471). The primary outcome was live birth rate. Secondary outcomes included positive pregnancy test rate, biochemical pregnancy rate, clinical pregnancy rate and miscarriage rate. Logistic regression models were performed, controlling for day of cryopreservation (day 5 or day 6), FET preparation, body mass index and oocyte age. A subanalysis was performed on embryos that underwent preimplantation genetic testing for aneuploidy (PGT-A). RESULTS:Unadjusted analysis showed no difference in live birth rate (42.5% versus 41.8% versus 46.9%; P = 0.167), pregnancy rate (58.8% versus 56.3% versus 62.4%; P = 0.085), biochemical pregnancy rate (8.9% versus 8.1% versus 8.5%; P = 0.827), clinical pregnancy rate (49.6% versus 48.1% versus 53.3%; P = 0.179), and spontaneous abortion rate (6.9% versus 6.0% versus 6.2%; P = 0.704) across the standard vitrification-standard warming, standard vitrification-fast warming, and fast vitrification-fast warming groups, respectively. These findings remained unchanged in the adjusted analysis. A subanalysis of PGT-A embryos showed no difference in live birth rate (45.1% versus 46.2% versus 49.6%; P = 0.472) between the standard vitrification-standard warming (n = 548), standard vitrification-fast warming (n = 424), and fast vitrification-fast warming (n = 266) groups, respectively. CONCLUSIONS:The fast vitrification-fast warming protocol for vitrified and warmed blastocyst-stage embryos is as effective as the standard vitrification-standard warming protocol, demonstrating no negative impact on live birth or secondary outcomes.

Sex-specific differences in fetal-placental signaling are well established. Female (F) fetuses favor glucocorticoid-regulated pathways, enhancing placental reserve but limiting growth. Male (M) fetuses prioritize androgen-driven signaling, promoting growth at the cost of adaptability. In PCOS, an androgen-mediated condition, these adaptations may be exaggerated, potentially altering placental histopathology. We retrospectively reviewed placental pathology data from singleton livebirths (1/2004–4/2022) conceived with FT (n = 1381). PCOS patients (Rotterdam criteria; n = 181) were grouped by fetal sex (M = 90; F = 91). Placental findings were categorized as anatomic, inflammatory, infectious, or vascular by a blinded perinatal pathologist using Amsterdam Workshop Consensus definitions. Comparisons were made using parametric/nonparametric tests. Logistic regression calculated crude and adjusted odds ratios (aOR), controlling for maternal age, BMI, race, gestational age, FT type, and gestational diabetes (GDM). Baseline characteristics of PCOS patients who delivered M vs F fetuses did not differ significantly [age, mean (SD)—M: 32.8 (3.1) vs F: 33.0 (3.8), p = 0.75; BMI— M: 26.2 (5.6) vs F: 26.2 (5.7), p = 0.89; nulliparity, n (%)—M: 67 (74.4%) vs F: 75 (83.3%), p = 0.14; FT—Ovulation Induction /Intrauterine Insemination—M: 40(44%) vs F: 48 (53.3%); In Vitro Fertilization—M: 51 (56%) vs F: 42 (46.7%), p = 0.21; GDM—M: 14 (15.4%) vs F: 8 (8.9%), p = 0.18]. There were no differences in anatomic, infectious, or vascular abnormalities by fetal sex in crude or adjusted models. Inflammatory abnormalities—villitis of unknown etiology, deciduitis, and intervillositis—were more frequent in F placentas on crude analysis [F: 20 (23%) vs M: 9 (9.9%), OR 1.14 (95% CI 1.02–1.27); p = 0.01], but not after adjustment [aOR 1.07 (95% CI 0.91–1.26); p = 0.4]. This study suggests that fetal sex does not significantly impact placental pathology in PCOS pregnancies. These findings provide reassurance that fetal sex differences may not be a major factor in placenta-mediated pregnancy complications in PCOS patients undergoing FT. Further research should explore the potential influence of androgen exposure on placental function and long-term fetal outcomes in this population.

Study question Can vision-based automated tracking algorithms reliably identify and track embryos throughout development? Summary answer AI-based EmID demonstrates high feasibility for embryo tracking, achieving 99.91% accuracy while traditional methods such as SIFT were less reliable with 89.82% overall accuracy. What is known already Ensuring accurate embryo identification in IVF is essential to prevent mix-ups that could lead to serious ethical and medical consequences. Traditional witnessing systems, such as manual verification and electronic tracking, monitor patient-level data but fail to track individual embryos throughout development. This limitation increases the risk of misidentification during culture, biopsy, and transfer. Vision-based, both traditional and AI-enabled, technologies offer potential solutions, but reliably distinguishing embryos remains challenging due to their continuous morphological changes. Biological variability poses significant hurdles for current automation technologies systems. Developing robust embryo-level witnessing methods can help in improving safety in IVF workflows for all stakeholders. Study design, size, duration The study was focused on evaluating the feasibility of vision-based tracking algorithms for embryo identification. Embryos were analyzed longitudinally from 5 hours post insemination (hpi) to 113 hpi, through assessments of embryo image pairs. Images captured using an Embryoscope at regular intervals were used to assess tracking performance over time. SIFT, a traditional feature-based method, and EmID, an AI-driven approach incorporating a vision transformer (ViT) architecture, were evaluated. Participants/materials, setting, methods EmID, a ViT-based siamese network, was developed for reliable embryo identification. It analyzes image pairs to determine embryo matches. Using 33 embryos, pairs were generated by matching each image (anchor) with a subsequent one at 10-minute to 1-hour intervals for temporal tracking. Positive pairs were from the same embryo at different times, while negative pairs were from different embryos at similar stages. SIFT used keypoint-based matching across these pairs. Main results and the role of chance We applied SIFT to identify each embryo at every time point among the set, comparing against every embryo at subsequent frames within a one-hour window. The overall identification accuracy was approximately 89.82% (SD: 4.49%; n = 4,264,661 embryo image pairs). Despite some accuracy drops at certain development points, our results confirmed that embryos can generally be tracked using visual features throughout development. However, its reliance on keypoint-based matching limited its robustness across developmental changes. In contrast, EmID, achieved 99.91% (SD: 0.12%; n = 4,264,661 embryo image pairs). When EmID’s performance at critical IVF time points (17, 60, and 110 hpi), corresponding to key stages like dish changes and embryo assessments (Fig. 9B). At these points, the likelihood of mix-ups is highest, making them crucial for clinical evaluation. EmID demonstrated perfected accuracy, sensitivity, specificity, PPV, and NPV at all three time points, demonstrating its potential for reliable embryo tracking in a clinical setting. Attention heatmaps demonstrated that EmID consistently focused on biologically relevant regions, including the zona pellucida, perivitelline space, and blastocoel, improving interpretability. These findings suggest that while traditional approaches can identify embryos to some extent, AI-driven models like EmID provide superior robustness and adaptability for embryo-level witnessing in IVF workflows. Limitations, reasons for caution The study was conducted in a controlled dataset, and with data collected from a single instrument. Additional validation in real-world clinical settings is required to confirm robustness across different laboratory conditions and instruments. Wider implications of the findings The feasibility of EmID for embryo tracking suggests its potential for integration into IVF workflows, reducing tracking errors and improving procedural reliability, which could lead to widespread adoption in embryology labs. Trial registration number No

How consistent are conventional AI (CAI) models in ranking embryos for IVF, and what impact does model variability have on clinical decision-making? CAI models exhibit significant ranking variability despite similar predictive performance, leading to inconsistent embryo selection and raising concerns about their reliability for clinical deployment. AI-driven embryo selection models primarily use Single Instance Learning (SIL), evaluating embryos independently to predict implantation or live birth. Despite high accuracy, SIL models exhibit significant ranking variability, with some performing no better than random chance, undermining trust in AI-driven IVF. This variability is often attributed to methodological differences rather than inherent model limitations. Consequently, the best-performing models, based on accuracy or area-under-the-curve, are selected for clinical use. However, this approach overlooks stability concerns, emphasizing the need for rigorous evaluations to ensure reliable, consistent embryo selection and prevent unpredictable patient outcomes in IVF. This study assessed AI model stability in embryo selection by training replicate SIL models on MGH embryo images. Models predicted live birth (LB) outcomes using identical architectures but varying random seeds. Rank orders were generated for patient cohorts (with ≥4 embryos) using softmax scores. Critical errors- ranking degenerate embryos as the most suitable for transfer despite the availability of blastocysts, were analyzed separately. Only Day 5 embryos were included; embryos with unknown outcomes were excluded. 50 replicate SIL models using ResNet50 were trained with Adam optimizer for 200 epochs, and with early stopping conditions in place. Data augmentation such as random flips, and rotations was utilized. With test sets of 92 patient cohorts from MGH and 49 from Cornell, rank orders were generated using softmax scores of each individual embryo in the embryo cohort. Kendall’s W measured ranking consistency, and critical error rates were calculated using the generated rank orders. SIL models exhibited substantial ranking variability, with Kendall’s W = 0.3571 ± 0.1302 (MGH) and 0.3410 ± 0.1398 (Cornell), indicating poor agreement across 50 replicate models trained on the same dataset. Critical error rates, where degenerate or arrested embryos were top ranked despite the presence of blastocysts, ranged from 3.61% to 21.69% (MGH) and 4.44% to 37.78% (Cornell), demonstrating inconsistent decision-making across sites. Conventional performance metrics (AUC, accuracy) failed to predict ranking stability, as SIL models achieved an average accuracy of 58.73% ± 4.58% and AUC of 60.02% ± 4.73% yet still produced discordant embryo rankings. Rank order inconsistency persisted regardless of dataset size, indicating instability arises from inherent model properties rather than data limitations. Models trained under identical conditions produced widely differing rankings due to variations in initialization seeds. At Cornell, error variance (68.59%) was three times higher than at MGH (22.53%), suggesting increased instability under distribution shifts. These findings highlight the risk that embryo selection outcomes could differ solely based on which SIL model is deployed, raising concerns about clinical reliability SIL models lack consistency, limiting their clinical applicability. More robust approaches, such as cohort-based learning, are needed to improve AI-driven embryo selection reliability. This study analyzed SIL model variability using data from two fertility centers, but findings may not generalize to other clinical settings. The absence of a definitive embryo ranking ground truth limits error validation. Additionally, model performance was assessed retrospectively; prospective validation is needed to confirm clinical impact. AI models for embryo selection require rigorous stability assessments beyond conventional accuracy metrics. SIL model variability highlights the need for alternative approaches, such as cohort-based learning, to enhance clinical reliability. Without robust validation, SIL-based embryo selection risks unpredictable clinical outcomes, undermining trust in CAI-driven IVF technologies. No

OBJECTIVE:To investigate the effect of supraphysiologic estradiol levels during ovarian stimulation on placental pathology among singleton livebirths conceived with in vitro fertilization (IVF) and fresh embryo transfer. DESIGN:Retrospective cohort study. SUBJECTS:Six hundred twenty six IVF-conceived singleton livebirths with associated placental pathology. EXPOSURE:Two separate analyses were performed, using the ≥75th (2588.75pg/mL), and ≥ 90th (3120.5 pg/mL) estradiol percentile as cut-offs. Livebirths were categorized as below or above the estradiol cut-off at time of trigger, and placental abnormalities were compared between groups. MAIN OUTCOME MEASURES:Placental pathology (abnormalities classified as: anatomic, vascular, infectious, and inflammatory). RESULTS:Mean (SD) age and body mass index were 35.2 (4.0) years and 24.8 (4.5) kg/m2. Most patients were identified as White (n = 482, 77.0%), and the most common infertility diagnosis was male factor (n = 178, 28.9%). Using the 75th percentile as estradiol cut-off, unadjusted rates of placental abnormalities did not differ between the two groups (anatomic: 63.7% vs. 65.9%; vascular: 72.0% vs. 66.1%; infectious:33.1% vs. 32%; and inflammatory: 14.6% vs. 20%, for ≥75th vs. <75th percentile, respectively). Adjusted risk ratio (adjRR) (95% confidence interval [CI]) was calculated to account for possible confounding factors (maternal age, body mass index, infertility diagnosis, stimulation protocol, gestational age at delivery, and infant sex). AdjRR (95% CI) did not show significant differences between groups in risk of anatomic or infectious abnormalities. In patients with estradiol levels ≥75th percentile, the risk of vascular abnormalities was higher vs. patients in the <75th percentile group (adjRR [95% CI] 1.14 [1.00-1.29]). Among patients with estradiol levels ≥75th percentile, the risk of inflammatory abnormalities was lower vs. patients in the <75th percentile group (adjRR [95% CI] 0.62 [0.38-0.99]). When using the 90th percentile as the cut-off, no significant differences were noted between groups in rates of anatomic, vascular, infectious, or inflammatory abnormalities (65.1% vs. 65.4%; 76.2% vs. 66.6%; 34.9% vs. 32%; and 15.9% vs. 19%, for ≥90th vs. <90th percentile, respectively). AdjRR (95% CI) did not show significant differences between groups in risk of anatomic, infectious, or inflammatory abnormalities. However, risk for vascular abnormalities was higher among patients with estradiol ≥90th vs. <90th percentile (adjRR [95% CI]: 1.19 [1.01-1.39]). CONCLUSION:Our study did not reveal clinically increased risks of placental abnormalities. Higher serum estradiol during IVF was associated with a marginally higher risk for vascular placental abnormalities.

OBJECTIVE:To evaluate the stability and reliability of artificial intelligence (AI) models and approaches in embryo selection and rank ordering for in vitro fertilization (IVF). DESIGN:A laboratory-based study evaluating the stability and consistency of single instance learning models that assess embryos individually, predicting live-birth outcomes based solely on each embryo's morphological features. Fifty replicate convolutional neural networks with varying initialization parameters were trained and tested across two independent fertility center datasets. Model performance was assessed through embryo rank ordering, critical error rates, and intermodel variability. Interpretability analyses using gradient-weighted class activation mapping and t-distributed stochastic neighbor embedding were conducted to explore decision-making discrepancies among replicate models. SUBJECTS:The study utilized retrospective embryo datasets from Massachusetts General Hospital and Weill Cornell Fertility Center, including images from 1,258 patients and 10,713 embryos from Massachusetts General Hospital, and 53 patients with 648 embryos from Cornell. MAIN OUTCOME MEASURES:Consistency in embryo ranking (Kendall's W), frequency of critical errors (instances where low-quality embryos were top-ranked), and intermodel variability across datasets. RESULTS:The AI models demonstrated poor consistency in embryo rank ordering (Kendall's W approximately 0.35) and exhibited high critical error rates (approximately 15%), often ranking lower-quality embryos above viable ones. Significant intermodel variability was observed even among models with similar predictive accuracies (area under curve approximately 60%). When tested on data from a different fertility center, model instability increased (error variance delta: 46.07%2), highlighting sensitivity to distribution shifts. Interpretability analyses revealed divergent decision-making strategies among replicate models, despite identical architectures and training protocols. CONCLUSION:Single instance learning AI models for IVF embryo selection exhibit substantial instability and inconsistency, undermining their clinical reliability. High intermodel variability and critical error rates raise concerns about their suitability for real-world deployment. This study highlights the need for more stable AI frameworks and robust evaluation metrics tailored to the clinical demands of IVF.

An international consensus meeting was convened to discuss globally applicable strategies for ‘future-proofing’ ART laboratories. The central theme was how the application of the foundational principles of laboratory accreditation enables any centre to create an ethos and framework that will support future-proofing in all regards. Discussions focussed on ART laboratory services from egg retrieval and semen specimen receipt to embryo transfer, as well as pertinent cryobanking activities. Issues related to whether ART treatment should be considered an essential service, overall clinic operations, general patient care, and the provision of clinical treatment, were not included as they fall under the purview of physicians and public health authorities. This report details the 16 core consensus points reached, which are supported by extensive practical recommendations that cover the gamut of ART laboratory operations.