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.

Self-organization of kinetochore-fibers in human mitotic spindles.

During eukaryotic cell division, chromosomes are linked to microtubules (MTs) in the spindle by a macromolecular complex called the kinetochore. The bound kinetochore microtubules (KMTs) are crucial to ensuring accurate chromosome segregation. Recent reconstructions by electron tomography (Kiewisz et al., 2022) captured the positions and configurations of every MT in human mitotic spindles, revealing that roughly half the KMTs in these spindles do not reach the pole. Here, we investigate the processes that give rise to this distribution of KMTs using a combination of analysis of large-scale electron tomography, photoconversion experiments, quantitative polarized light microscopy, and biophysical modeling. Our results indicate that in metaphase, KMTs grow away from the kinetochores along well-defined trajectories, with the speed of the KMT minus ends continually decreasing as the minus ends approach the pole, implying that longer KMTs grow more slowly than shorter KMTs. The locations of KMT minus ends, and the turnover and movements of tubulin in KMTs, are consistent with models in which KMTs predominately nucleate de novo at kinetochores in metaphase and are inconsistent with substantial numbers of non-KMTs being recruited to the kinetochore in metaphase. Taken together, this work leads to a mathematical model of the self-organization of kinetochore-fibers in human mitotic spindles.

Mechanical Mechanisms of Chromosome Segregation.

Chromosome segregation-the partitioning of genetic material into two daughter cells-is one of the most crucial processes in cell division. In all Eukaryotes, chromosome segregation is driven by the spindle, a microtubule-based, self-organizing subcellular structure. Extensive research performed over the past 150 years has identified numerous commonalities and contrasts between spindles in different systems. In this review, we use simple coarse-grained models to organize and integrate previous studies of chromosome segregation. We discuss sites of force generation in spindles and fundamental mechanical principles that any understanding of chromosome segregation must be based upon. We argue that conserved sites of force generation may interact differently in different spindles, leading to distinct mechanical mechanisms of chromosome segregation. We suggest experiments to determine which mechanical mechanism is operative in a particular spindle under study. Finally, we propose that combining biophysical experiments, coarse-grained theories, and evolutionary genetics will be a productive approach to enhance our understanding of chromosome segregation in the future.

Freezing on a sphere.

The best understood crystal ordering transition is that of two-dimensional freezing, which proceeds by the rapid eradication of lattice defects as the temperature is lowered below a critical threshold. But crystals that assemble on closed surfaces are required by topology to have a minimum number of lattice defects, called disclinations, that act as conserved topological charges-consider the 12 pentagons on a football or the 12 pentamers on a viral capsid. Moreover, crystals assembled on curved surfaces can spontaneously develop additional lattice defects to alleviate the stress imposed by the curvature. It is therefore unclear how crystallization can proceed on a sphere, the simplest curved surface on which it is impossible to eliminate such defects. Here we show that freezing on the surface of a sphere proceeds by the formation of a single, encompassing crystalline 'continent', which forces defects into 12 isolated 'seas' with the same icosahedral symmetry as footballs and viruses. We use this broken symmetry-aligning the vertices of an icosahedron with the defect seas and unfolding the faces onto a plane-to construct a new order parameter that reveals the underlying long-range orientational order of the lattice. The effects of geometry on crystallization could be taken into account in the design of nanometre- and micrometre-scale structures in which mobile defects are sequestered into self-ordered arrays. Our results may also be relevant in understanding the properties and occurrence of natural icosahedral structures such as viruses.

Phase behavior of charged colloids at a fluid interface.

We study the phase behavior of a system of charged colloidal particles that are electrostatically bound to an almost flat interface between two fluids. We show that, despite the fact that our experimental system consists of only 10^{3}-10^{4} particles, the phase behavior is consistent with the theory of melting due to Kosterlitz, Thouless, Halperin, Nelson, and Young. Using spatial and temporal correlations of the bond-orientational order parameter, we classify our samples into solid, isotropic fluid, and hexatic phases. We demonstrate that the topological defect structure we observe in each phase corresponds to the predictions of Kosterlitz-Thouless-Halperin-Nelson-Young theory. By measuring the dynamic Lindemann parameter γ_{L}(τ) and the non-Gaussian parameter α_{2}(τ) of the displacements of the particles relative to their neighbors, we show that each of the phases displays distinctive dynamical behavior.

Long-wavelength fluctuations and the glass transition in two dimensions and three dimensions.

Phase transitions significantly differ between 2D and 3D systems, but the influence of dimensionality on the glass transition is unresolved. We use microscopy to study colloidal systems as they approach their glass transitions at high concentrations and find differences between two dimensions and three dimensions. We find that, in two dimensions, particles can undergo large displacements without changing their position relative to their neighbors, in contrast with three dimensions. This is related to Mermin-Wagner long-wavelength fluctuations that influence phase transitions in two dimensions. However, when measuring particle motion only relative to their neighbors, two dimensions and three dimensions have similar behavior as the glass transition is approached, showing that the long-wavelength fluctuations do not cause a fundamental distinction between 2D and 3D glass transitions.

Charged hydrophobic colloids at an oil-aqueous phase interface.

Hydrophobic poly(methyl methacrylate) (PMMA) colloidal particles, when dispersed in oil with a relatively high dielectric constant, can become highly charged. In the presence of an interface with a conducting aqueous phase, image-charge effects lead to strong binding of colloidal particles to the interface, even though the particles are wetted very little by the aqueous phase. We study both the behavior of individual colloidal particles as they approach the interface and the interactions between particles that are already interfacially bound. We demonstrate that using particles which are minimally wetted by the aqueous phase allows us to isolate and study those interactions which are due solely to charging of the particle surface in oil. Finally, we show that these interactions can be understood by a simple image-charge model in which the particle charge q is the sole fitting parameter.