BlastAssist: a deep learning pipeline to measure interpretable features of human embryos.

STUDY QUESTION: Can the BlastAssist deep learning pipeline perform comparably to or outperform human experts and embryologists at measuring interpretable, clinically relevant features of human embryos in IVF? SUMMARY ANSWER: The BlastAssist pipeline can measure a comprehensive set of interpretable features of human embryos and either outperform or perform comparably to embryologists and human experts in measuring these features. WHAT IS KNOWN ALREADY: Some studies have applied deep learning and developed 'black-box' algorithms to predict embryo viability directly from microscope images and videos but these lack interpretability and generalizability. Other studies have developed deep learning networks to measure individual features of embryos but fail to conduct careful comparisons to embryologists' performance, which are fundamental to demonstrate the network's effectiveness. STUDY DESIGN, SIZE, DURATION: We applied the BlastAssist pipeline to 67 043 973 images (32 939 embryos) recorded in the IVF lab from 2012 to 2017 in Tel Aviv Sourasky Medical Center. We first compared the pipeline measurements of individual images/embryos to manual measurements by human experts for sets of features, including: (i) fertilization status (n = 207 embryos), (ii) cell symmetry (n = 109 embryos), (iii) degree of fragmentation (n = 6664 images), and (iv) developmental timing (n = 21 036 images). We then conducted detailed comparisons between pipeline outputs and annotations made by embryologists during routine treatments for features, including: (i) fertilization status (n = 18 922 embryos), (ii) pronuclei (PN) fade time (n = 13 781 embryos), (iii) degree of fragmentation on Day 2 (n = 11 582 embryos), and (iv) time of blastulation (n = 3266 embryos). In addition, we compared the pipeline outputs to the implantation results of 723 single embryo transfer (SET) cycles, and to the live birth results of 3421 embryos transferred in 1801 cycles. PARTICIPANTS/MATERIALS, SETTING, METHODS: In addition to EmbryoScope™ image data, manual embryo grading and annotations, and electronic health record (EHR) data on treatment outcomes were also included. We integrated the deep learning networks we developed for individual features to construct the BlastAssist pipeline. Pearson's χ2 test was used to evaluate the statistical independence of individual features and implantation success. Bayesian statistics was used to evaluate the association of the probability of an embryo resulting in live birth to BlastAssist inputs. MAIN RESULTS AND THE ROLE OF CHANCE: The BlastAssist pipeline integrates five deep learning networks and measures comprehensive, interpretable, and quantitative features in clinical IVF. The pipeline performs similarly or better than manual measurements. For fertilization status, the network performs with very good parameters of specificity and sensitivity (area under the receiver operating characteristics (AUROC) 0.84-0.94). For symmetry score, the pipeline performs comparably to the human expert at both 2-cell (r = 0.71 ± 0.06) and 4-cell stages (r = 0.77 ± 0.07). For degree of fragmentation, the pipeline (acc = 69.4%) slightly under-performs compared to human experts (acc = 73.8%). For developmental timing, the pipeline (acc = 90.0%) performs similarly to human experts (acc = 91.4%). There is also strong agreement between pipeline outputs and annotations made by embryologists during routine treatments. For fertilization status, the pipeline and embryologists strongly agree (acc = 79.6%), and there is strong correlation between the two measurements (r = 0.683). For degree of fragmentation, the pipeline and embryologists mostly agree (acc = 55.4%), and there is also strong correlation between the two measurements (r = 0.648). For both PN fade time (r = 0.787) and time of blastulation (r = 0.887), there's strong correlation between the pipeline and embryologists. For SET cycles, 2-cell time (P < 0.01) and 2-cell symmetry (P < 0.03) are significantly correlated with implantation success rate, while other features showed correlations with implantation success without statistical significance. In addition, 2-cell time (P < 5 × 10-11), PN fade time (P < 5 × 10-10), degree of fragmentation on Day 3 (P < 5 × 10-4), and 2-cell symmetry (P < 5 × 10-3) showed statistically significant correlation with the probability of the transferred embryo resulting in live birth. LIMITATIONS, REASONS FOR CAUTION: We have not tested the BlastAssist pipeline on data from other clinics or other time-lapse microscopy (TLM) systems. The association study we conducted with live birth results do not take into account confounding variables, which will be necessary to construct an embryo selection algorithm. Randomized controlled trials (RCT) will be necessary to determine whether the pipeline can improve success rates in clinical IVF. WIDER IMPLICATIONS OF THE FINDINGS: BlastAssist provides a comprehensive and holistic means of evaluating human embryos. Instead of using a black-box algorithm, BlastAssist outputs meaningful measurements of embryos that can be interpreted and corroborated by embryologists, which is crucial in clinical decision making. Furthermore, the unprecedentedly large dataset generated by BlastAssist measurements can be used as a powerful resource for further research in human embryology and IVF. STUDY FUNDING/COMPETING INTEREST(S): This work was supported by Harvard Quantitative Biology Initiative, the NSF-Simons Center for Mathematical and Statistical Analysis of Biology at Harvard (award number 1764269), the National Institute of Heath (award number R01HD104969), the Perelson Fund, and the Sagol fund for embryos and stem cells as part of the Sagol Network. The authors declare no competing interests. TRIAL REGISTRATION NUMBER: Not applicable.

The impact of fragile X premutation carrier status on embryo morphokinetic development.

RESEARCH QUESTION: Does inheritance of the fragile X mental retardation 1 (FMR1) premutation allele affect embryo morphokinetic development? DESIGN: A retrospective cohort analysis of 529 embryos from 126 IVF cycles of 39 FMR1 premutation female carriers undergoing preimplantation genetic testing for monogenic/single gene defects (PGT-M). Morphological and morphokinetic parameters obtained using a time-lapse monitoring system were compared between embryos that inherited the FMR1 premutation allele (FMR1 group, n = 271) and those who received the normal allele (normal group, n = 258). The following embryo outcome measures were compared: morphokinetic parameters up to day 3, start of blastulation time (tSB) for day 5 embryos and the rate of top-quality embryos on days 3 and 5. RESULTS: No differences were found in morphokinetic parameters between the groups from the time of intracytoplasmic sperm injection (ICSI) until a biopsy on day 3. The blastulation rate in the two groups was comparable. However, the start of blastulation was delayed in FMR1 embryos compared to that in the genetically normal embryos (median tSB: 104.2 h [99.3-110.3] versus 101.6 h [94.5-106.7], P = 0.01). In addition, the rate of top-quality FMR1 embryos was lower than that of genetically normal embryos (25.6% versus 38.8%, P = 0.04). CONCLUSION: Embryos that inherit the FMR1 premutation allele are of lower quality at the blastocyst stage compared with those that do not inherit the mutated allele.

Meta-analysis of real-time fMRI neurofeedback studies using individual participant data: How is brain regulation mediated?

An increasing number of studies using real-time fMRI neurofeedback have demonstrated that successful regulation of neural activity is possible in various brain regions. Since these studies focused on the regulated region(s), little is known about the target-independent mechanisms associated with neurofeedback-guided control of brain activation, i.e. the regulating network. While the specificity of the activation during self-regulation is an important factor, no study has effectively determined the network involved in self-regulation in general. In an effort to detect regions that are responsible for the act of brain regulation, we performed a post-hoc analysis of data involving different target regions based on studies from different research groups. We included twelve suitable studies that examined nine different target regions amounting to a total of 175 subjects and 899 neurofeedback runs. Data analysis included a standard first- (single subject, extracting main paradigm) and second-level (single subject, all runs) general linear model (GLM) analysis of all participants taking into account the individual timing. Subsequently, at the third level, a random effects model GLM included all subjects of all studies, resulting in an overall mixed effects model. Since four of the twelve studies had a reduced field of view (FoV), we repeated the same analysis in a subsample of eight studies that had a well-overlapping FoV to obtain a more global picture of self-regulation. The GLM analysis revealed that the anterior insula as well as the basal ganglia, notably the striatum, were consistently active during the regulation of brain activation across the studies. The anterior insula has been implicated in interoceptive awareness of the body and cognitive control. Basal ganglia are involved in procedural learning, visuomotor integration and other higher cognitive processes including motivation. The larger FoV analysis yielded additional activations in the anterior cingulate cortex, the dorsolateral and ventrolateral prefrontal cortex, the temporo-parietal area and the visual association areas including the temporo-occipital junction. In conclusion, we demonstrate that several key regions, such as the anterior insula and the basal ganglia, are consistently activated during self-regulation in real-time fMRI neurofeedback independent of the targeted region-of-interest. Our results imply that if the real-time fMRI neurofeedback studies target regions of this regulation network, such as the anterior insula, care should be given whether activation changes are related to successful regulation, or related to the regulation process per se. Furthermore, future research is needed to determine how activation within this regulation network is related to neurofeedback success.

Cigarette smoking leads to persistent and dose-dependent alterations of brain activity and connectivity in anterior insula and anterior cingulate.

Although many smokers try to quit smoking, only about 20-25 percent will achieve abstinence despite 6 months or more of gold-standard treatment. This low success rate suggests long-term changes in the brain related to smoking, which remain poorly understood. We compared ex-smokers to both active smokers and non-smokers using functional magnetic resonance imaging (fMRI) to explore persistent modifications in brain activity and network organization. This prospective and consecutive study includes 18 non-smokers (29.5 ± 6.7 years of age, 11 women), 14 smokers (≥10 cigarettes a day >2 years of smoking, 29.3 ± 6.0 years of age, 10 women) and 14 ex-smokers (>1 year of quitting 30.5 ± 5.7 years of age, 10 women). Participants underwent a block-design fMRI study contrasting smoking cue with control (neutral cue) videos. Data analyses included task-related general linear model, seed-based functional connectivity, voxel-based morphometry (VBM) of gray matter and tract-based spatial statistics (TBSS) of white matter. Smoking cue videos versus control videos activated the right anterior insula in ex-smokers compared with smokers, an effect correlating with cumulative nicotine intake (pack-years). Moreover, ex-smokers had a persistent decrease in functional connectivity between right anterior insula and anterior cingulate cortex (ACC) compared with control participants, but similar to active smokers. Potentially confounding alterations in gray or white matter were excluded in VBM and TBSS analyses. In summary, ex-smokers with long-term nicotine abstinence have persistent and dose-dependent brain network changes notably in the right anterior insula and its connection to the ACC.

Dynamic reconfiguration of human brain functional networks through neurofeedback.

Recent fMRI studies demonstrated that functional connectivity is altered following cognitive tasks (e.g., learning) or due to various neurological disorders. We tested whether real-time fMRI-based neurofeedback can be a tool to voluntarily reconfigure brain network interactions. To disentangle learning-related from regulation-related effects, we first trained participants to voluntarily regulate activity in the auditory cortex (training phase) and subsequently asked participants to exert learned voluntary self-regulation in the absence of feedback (transfer phase without learning). Using independent component analysis (ICA), we found network reconfigurations (increases in functional network connectivity) during the neurofeedback training phase between the auditory target region and (1) the auditory pathway; (2) visual regions related to visual feedback processing; (3) insula related to introspection and self-regulation and (4) working memory and high-level visual attention areas related to cognitive effort. Interestingly, the auditory target region was identified as the hub of the reconfigured functional networks without a-priori assumptions. During the transfer phase, we again found specific functional connectivity reconfiguration between auditory and attention network confirming the specific effect of self-regulation on functional connectivity. Functional connectivity to working memory related networks was no longer altered consistent with the absent demand on working memory. We demonstrate that neurofeedback learning is mediated by widespread changes in functional connectivity. In contrast, applying learned self-regulation involves more limited and specific network changes in an auditory setup intended as a model for tinnitus. Hence, neurofeedback training might be used to promote recovery from neurological disorders that are linked to abnormal patterns of brain connectivity.

Recovery of the default mode network after demanding neurofeedback training occurs in spatio-temporally segregated subnetworks.

The default mode (DM) network is a major large-scale cerebral network that can be identified with functional magnetic resonance imaging (fMRI) during resting state. Most studies consider functional connectivity networks as stationary phenomena. Consequently, the transient behavior of the DM network and its subnetworks is still largely unexplored. Most functional connectivity fMRI studies assess the steady state of resting without any task. To specifically investigate the recovery of the DM network during the transition from activation to rest, we implemented a cognitively demanding real-time fMRI neurofeedback task that targeted down-regulation of the primary auditory cortex. Each of twelve healthy subjects performed 16 block-design fMRI runs (4 runs per day repeated on 4 days) resulting 192 runs in total. The analysis included data-driven independent component analysis (ICA) and high-resolution latency estimation between the four components that corresponded to subnetworks of the DM network. These different subnetworks reemerged after regulation with an average time lag or 3.3s and a time lag of 4.4s between the first and fourth components; i.e., the DM recovery first shifts from anterior to posterior, and then gradually focuses on the ventral part of the posterior cingulate cortex, which is known to be implicated in internally directed cognition. In addition, we found less reactivation in the early anterior subnetwork as regulation strength increased, but more reactivation with larger regulation for the late subnetwork that encompassed the ventral PCC. This finding confirms that the level of task engagement influences inversely the subsequent recovery of regions related to attention compared to those related to internally directed cognition.

Transfer of Day 6 Frozen-Thawed Blastocysts on Day 5 Compared with Day 6: Catching Up with the Window of Implantation-a Retrospective Study.

To compare clinical pregnancy rate (CPR) and live birth rate (LBR) after frozen-thawed embryo transfer (FET) of day (D-) 6 blastocysts on D-5 versus D-6. A retrospective cohort study. A university-affiliated single-center tertiary hospital. Women undergoing FET of D-6 blastocysts between August 2015 and March 2019 were included in the study. Exclusion criteria were endometriosis and maternal age ≥ 42. Cycles involving embryo transfer (ET) at D-6 were compared to cycles involving ET on D-5. Primary outcomes assessed were CPR and LBR, and the secondary outcomes were spontaneous abortion and chemical pregnancy rates. Forty-two cycles were assessed, 21 in which ET occurred on D-6 and 21 in which ET occurred on D-5. There were no significant differences between groups regarding age, body mass index (BMI), etiology of infertility, number of oocytes aspirated and blastocysts cryopreserved in the fresh cycle, reason for freezing on D-6, endometrial thickness before ET, and blastocyst grade. A comparison of outcomes of ET on D-5 with those involving ET on D-6 revealed that D-5 transfer produced significantly higher CPR (8, 38% vs. 2, 8.5%; P = 0.030) and LBR (6, 28.6% vs. 1, 4.8%; P = 0.038), respectively. FET of D-6 embryos on D-5 compared with D-6 is associated with increased CPR and LBR values. These findings might be related to the limited time window for optimal rates of implantation and indicate that transferring embryos on D-6 of a FET cycle is likely too late.

Distributed Patterns of Brain Activity Underlying Real-Time fMRI Neurofeedback Training.

Neurofeedback (NF) based on real-time functional magnetic resonance imaging (rt-fMRI) is an exciting neuroimaging application. In most rt-fMRI NF studies, the activity level of a single region of interest (ROI) is provided as a feedback signal and the participants are trained to up or down regulate the feedback signal. NF training effects are typically analyzed using a confirmatory univariate approach, i.e., changes in the target ROI are explained by a univariate linear modulation. However, learning to self-regulate the ROI activity through NF is mediated by distributed changes across the brain. Here, we deploy a multivariate decoding model for assessing NF training effects across the whole brain. Specifically, we first explain the NF training effect by a posthoc multivariate model that leads to a pattern of coactivation based on 90 functional atlas regions. We then use cross validation to reveal the set of brain regions with the best fit. This novel approach was applied to the data from a rt-fMRI NF study where the participants learned to down regulate the auditory cortex. We found that the optimal model consisted of 16 brain regions whose coactivation patterns best described the training effect over the NF training days. Cross validation of the multivariate model showed that it generalized across the participants. Interestingly, the participants could be clustered into two groups with distinct patterns of coactivation, potentially reflecting different NF learning strategies. Overall, our findings revealed that multiple brain regions are involved in learning to regulate an activity in a single ROI, and thus leading to a better understanding of the mechanisms underlying NF training.

Continuous vs. intermittent neurofeedback to regulate auditory cortex activity of tinnitus patients using real-time fMRI - A pilot study.

The emerging technique of real-time fMRI neurofeedback trains individuals to regulate their own brain activity via feedback from an fMRI measure of neural activity. Optimum feedback presentation has yet to be determined, particularly when working with clinical populations. To this end, we compared continuous against intermittent feedback in subjects with tinnitus. Fourteen participants with tinnitus completed the whole experiment consisting of nine runs (3 runs × 3 days). Prior to the neurofeedback, the target region was localized within the auditory cortex using auditory stimulation (1 kHz tone pulsating at 6 Hz) in an ON-OFF block design. During neurofeedback runs, participants received either continuous (n = 7, age 46.84 ± 12.01, Tinnitus Functional Index (TFI) 49.43 ± 15.70) or intermittent feedback (only after the regulation block) (n = 7, age 47.42 ± 12.39, TFI 49.82 ± 20.28). Participants were asked to decrease auditory cortex activity that was presented to them by a moving bar. In the first and the last session, participants also underwent arterial spin labeling (ASL) and resting-state fMRI imaging. We assessed tinnitus severity using the TFI questionnaire before all sessions, directly after all sessions and six weeks after all sessions. We then compared neuroimaging results from neurofeedback using a general linear model (GLM) and region-of-interest analysis as well as behavior measures employing a repeated-measures ANOVA. In addition, we looked at the seed-based connectivity of the auditory cortex using resting-state data and the cerebral blood flow using ASL data. GLM group analysis revealed that a considerable part of the target region within the auditory cortex was significantly deactivated during neurofeedback. When comparing continuous and intermittent feedback groups, the continuous group showed a stronger deactivation of parts of the target region, specifically the secondary auditory cortex. This result was confirmed in the region-of-interest analysis that showed a significant down-regulation effect for the continuous but not the intermittent group. Additionally, continuous feedback led to a slightly stronger effect over time while intermittent feedback showed best results in the first session. Behaviorally, there was no significant effect on the total TFI score, though on a descriptive level TFI scores tended to decrease after all sessions and in the six weeks follow up in the continuous group. Seed-based connectivity with a fixed-effects analysis revealed that functional connectivity increased over sessions in the posterior cingulate cortex, premotor area and part of the insula when looking at all patients while cerebral blood flow did not change significantly over time. Overall, these results show that continuous feedback is suitable for long-term neurofeedback experiments while intermittent feedback presentation promises good results for single session experiments when using the auditory cortex as a target region. In particular, the down-regulation effect is more pronounced in the secondary auditory cortex, which might be more susceptible to voluntary modulation in comparison to a primary sensory region.

Neural Correlates of Clinical Scores in Patients with Anterior Shoulder Apprehension.

INTRODUCTION: Anterior shoulder apprehension is a commonly reported complaint in anterior shoulder instability, which may lead to patient morbidity and impede shoulder function. It is the result of a cognitively complex mechanism, which includes anxiety, salience, fear, and anticipation. PURPOSE: The aim of this prospective case-control study was to correlate five clinically established scores using functional magnetic resonance imaging to assess brain activation patterns in patients with apprehension related to anterior shoulder instability. METHODS: This study includes 28 consecutive male right-handed patients (mean ± SEM, 26.8 ± 1.2 yr) with positive shoulder apprehension test and 10 healthy matched control participants without apprehension or a history of instability. Task-related and functional connectivity functional magnetic resonance imaging activation patterns occurring during apprehension video cue stimulation were correlated with five clinical tests and scores: Visual Analog Scale (VAS), Rowe score for instability, Simple Shoulder Test, Subjective Shoulder Value (SSV), and Western Ontario Shoulder Instability (WOSI). RESULTS: Rowe, pain VAS, and WOSI scores correlated with prefrontal cortex, dorsolateral prefrontal cortex, dorsomedial prefrontal cortex, somatosensory area, and parieto-occipital and temporal areas (default mode network). Rowe score additionally correlated with frontal pole, anterior midcingulate cortex, and visual areas. Moreover, SSV correlated with task-related brain activity in the bilateral precentral gyrus, bilateral postcentral gyrus, and bilateral superior parietal lobe. CONCLUSIONS: Overall, Rowe score provides the strongest link between shoulder apprehension and brain level alterations as it correlates with the highest number of independent components involving areas responsible for both motor and cognitive functions, whereas pain VAS and WOSI occupy an intermediately strong link recruiting less brain networks. Finally, Simple Shoulder Test and SSV have the weakest link at the brain level.