Pith. sign in

REVIEW 3 major objections 6 minor 26 references

Time-Lapse Video-Based Embryo Grading via Complementary Spatial-Temporal Pattern Mining

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper introduces video-based embryo grading as a new task, curates a 2,596-video clinical time-lapse dataset with embryologist quality labels, and proposes CoSTeM, a dual-branch network that mines complementary morphological and…

desk verdict First holistic video-based embryo grading benchmark, but the headline SOTA claim over AIM is not statistically supported by a single validation split. read the letter →

arxiv 2506.04950 v1 pith:VTQJVEOL submitted 2025-06-05 cs.CV

Yong Sun , Yipeng Wang , Junyu Shi , Zhiyuan Zhang , Yanmei Xiao , Lei Zhu , Manxi Jiang , Qiang Nie This is my paper · ORCID
classification cs.CV
keywords video-basedembryogradingtime-lapsemonitoringin-vitrofertilizationqualityassessmentmorphokineticanalysismixtureofcross-attentiveexpertstemporaltransformer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish a new benchmark task: automatically grading IVF embryo quality directly from full-length time-lapse videos, rather than from isolated images or from clinical outcomes confounded by maternal and uterine factors. It curates a 2,596-video clinical dataset annotated by embryologists into poor/fair/good grades, and proposes CoSTeM, a dual-branch framework that separately mines static morphology and dynamic morphokinetics before combining them for classification. The authors report that CoSTeM outperforms prior embryo-selection methods and general video-action-recognition models, reaching 0.8606 accuracy and 0.7618 macro F1 on their validation set. If the approach holds, clinics could prioritize embryos for transfer with less manual video review, and the task would give the field a common benchmark aligned with daily clinical practice.

What carries the argument

The load-bearing mechanism is the complementary decomposition of each video into a spatial pattern stream and a temporal pattern stream, recombined only at the classification head. The morphological branch operates on per-frame patch tokens and applies, per frame, a Mixture of Cross-Attentive Experts: $M=8$ learnable expert vectors, each attending to the frame's tokens as a query, with an MLP router weighting the experts by estimated developmental stage, producing one condensed frame feature. A Temporal Selection Block then uses $K=8$ learnable temporal expert queries that cross-attend over the frame-feature sequence, and a diversity loss pushes these experts to select distinct frames. In parallel, the morphokinetic branch feeds the per-frame [CLS] tokens through a two-block Temporal Transformer to encode the full developmental trajectory; the two streams' outputs are concatenated along the channel dimension and passed to an MLP classifier. A frozen pretrained image encoder with lightweight MLP adaptors supplies hierarchical frame features.

What would settle it

Re-annotate a random subset of the 2,596 videos by two or more trained embryologists using the same protocol; if inter-rater agreement is low (say Cohen's kappa below 0.4), the reported accuracy and F1 measure agreement with one noisy reference rather than true embryo quality. A second test: run CoSTeM on time-lapse videos from another clinic or time-lapse monitoring device; if accuracy drops by more than a few points, the single-center dataset cannot support the general claim.

Watch

Extended reading notes

Core claim

The central discovery claimed is that a single network can learn both the 'what' and the 'when' of embryo quality from one continuous video: local morphological signs (cell structure, fragmentation, stage-specific features) and global developmental timing (cleavage dynamics, pronucleus appearance, blastocyst formation) are complementary, and their joint modeling reproduces embryologists' holistic grading better than either alone. On the curated 2,596-video dataset, the complete CoSTeM reaches 0.8606 accuracy, 0.7746 precision, 0.7595 recall, and 0.7618 macro-averaged F1, the best among compared embryo-selection and action-recognition methods; the closest competitor ties on accuracy but trails on precision, recall, and F1. Ablations show that removing either branch or the selection mechanisms degrades performance, and that the learned temporal experts concentrate on pronucleus, cleavage, and blastocyst stages, matching standard embryologist protocols.

Load-bearing premise

The load-bearing premise is that the embryologists' holistic grade labels are consistent enough to train on: the paper does not report how many embryologists annotated, what the grading protocol was in detail, or how much the graders agreed, and if the labels are noisy the model's accuracy inherits that noise.

Editorial extensions

If this is right

  • If the result holds, a clinic's time-lapse system could output a three-grade holistic quality score for every embryo with no additional manual review, letting embryologists concentrate on ambiguous cases.
  • The task definition—predict the embryologist's global grade rather than a clinical outcome—avoids confounding from uterine environment and maternal factors that plagued outcome-based models, so the reported numbers reflect video information only.
  • The temporal experts' learned preference for pronucleus, cleavage, and blastocyst frames suggests the model is keyed to the same developmental milestones that guide human grading, which supports explainability.
  • Because 64 uniformly sampled frames at 2-hour intervals are used, the method could run on a fixed temporal grid, making deployment in existing time-lapse monitoring systems straightforward.
  • If the dataset is released as promised, it becomes a shared benchmark on which future embryo-grading and video-understanding methods can be compared directly.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the reported 0.8606 accuracy is probably optimistic relative to real-world deployment because all 2,596 videos come from one clinic and one time-lapse system; a multi-center test with varied hardware and patient demographics is the natural next experiment, and the current numbers set a ceiling, not an expectation.
  • Editorial inference: since the ground truth is embryologists' subjective grades and the paper does not report inter-observer agreement, the same architecture could be evaluated against a consensus-grade subset or a re-annotation study to separate model skill from label noise.
  • Editorial inference: the complementary-branch design is general; the same spatial/temporal decomposition could be tested on other long medical videos where local structural detail and global progression both matter, such as gastroscopy or embryo development in other species.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper introduces Video-Based Embryo Grading, a three-class (poor/fair/good) task of predicting embryologists' holistic quality assessments from full-length time-lapse microscopy videos. The authors curate a 2,596-video clinical dataset from an EmbryoScope system, propose the CoSTeM framework consisting of a morphological branch (Mixture of Cross-Attentive Experts plus Temporal Selection Block) and a morphokinetic branch (Temporal Transformer) built on a frozen CLIP ViT with a bypass adaptation network, and report comparisons against embryo-selection and action-recognition baselines plus ablations. The claimed result is state-of-the-art performance, e.g., accuracy 0.8606 and macro F1 0.7618 on a single 70/30 split.

Significance. If the result holds, the paper makes a useful contribution by defining a clinically aligned task, releasing a real clinical dataset (on acceptance), and proposing a plausible two-branch architecture that mimics morphological and morphokinetic assessment. The comparison protocol uses official codebases or specifications, and the ablations consistently favor CoSTeM, which lends internal credibility. The main value is as a benchmark and baseline for a task that appears underexplored; however, the empirical support for the 'best performance' claim currently rests on a single split, so the significance hinges on the strength of the statistical evidence and the reliability of the clinical labels.

major comments (3)
  1. [§4.1, Tables 1-3] All quantitative claims are single-run point estimates on one 70/30 split, with the validation set used for checkpoint selection; no standard deviations, confidence intervals, significance tests, or repeated runs are reported. In Table 1, CoSTeM and AIM have identical accuracy (0.8606) and the F1 margin is 1.41 percentage points (0.7618 vs 0.7477), corresponding to roughly 10-15 validation videos; this margin is of the same order as seed-level variation for transformer-based video models. The central claim that CoSTeM 'achieves the best performance' is therefore not statistically established. Please add multi-seed runs with mean and standard deviation and paired significance tests, or evaluate on a separate test set not used for model selection.
  2. [§2] The ground-truth annotation is described only as a 'systematic, multi-stage evaluation process' by embryologists; the manuscript does not report the number of annotators, inter-observer agreement (e.g., Cohen's or Fleiss' kappa), or the precise grading protocol used. Since the labels are subjective holistic grades, label noise and bias propagate directly into the reported accuracy, and the benchmark's reliability cannot be assessed without this information. Please document the annotation procedure and reliability statistics, and state whether labels were adjudicated.
  3. [§2 and §4.1] All videos come from a single IVF center (anonymized as '∗ ∗ ∗Hospital') using one EmbryoScope system, and no external validation is performed. This leaves generalization across clinics, incubator models, and patient populations untested; for a 'foundational benchmark' and a clinical-translation claim, single-center evaluation is a significant limitation. Please add external validation on an independent cohort, or explicitly restrict the claims to the single-center setting.
minor comments (6)
  1. [§3.2] The router weights W_i are described as approximating probabilities of developmental-stage membership, but no normalization (e.g., softmax) is specified; please state how W_i is computed and constrained.
  2. [§4.3, Fig. 3(b)] The claim that the temporal experts 'align perfectly' with pronucleus formation, cleavage, and blastocyst stages is qualitative; please overlay expert-annotated stage boundaries or report a quantitative agreement metric.
  3. [§4.1] Uniform sampling at 2-hour intervals between 16 and 142 hours yields up to 64 frames, but the extent of zero-padding and its effect on the morphokinetic branch is not reported; please provide statistics on how many videos have missing frames at sampled time points.
  4. [Tables 1-3] Metrics are reported to four decimals on a validation set of about 780 videos; consider reporting confidence intervals or rounding to a meaningful precision.
  5. [§4.1] For the baselines implemented from official codebases or paper specifications, please list any protocol differences (e.g., input frame counts, input resolution, or sampling strategy) that could affect the fairness of the comparison.
  6. [Figure 1 caption] The caption contains a typo: 'a embryo' should be 'an embryo'.

Circularity Check

1 steps flagged · score 4.0 of 10

Reported 'best performance' is the optimal validation-set checkpoint, so the headline comparison is partly selected rather than independently predicted; the architecture itself relies on external components and no self-citation chain.

  1. fitted input called prediction [Section 4.1 'Evaluation Metrics'; Section 4.2 'Comparisons with Advanced Methods', Table 1]
    "We trained models on the training set and evaluated their optimal performances on the validation set using Accuracy, Precision, Recall, and F1-score."

    The phrase 'optimal performances on the validation set' means the checkpoint and hyperparameters are selected on the validation metric that the paper then reports as the method's result. No separate test set and no repeated splits are reported, so Table 1's accuracy/F1 are the maximum over training epochs on the very set used for model selection, not a prediction on data that never influenced the model choice. The claimed advantage over AIM rests on a 1.41-point macro-F1 margin (0.7618 vs 0.7477) selected in this way; hence the headline 'best performance' is partly an artifact of the selection criterion, not an independent held-out comparison.

full rationale

The paper's methodological building blocks (CLIP pretraining, ViViT-style temporal transformer, DETR-style cross-attention, Mix-of-Experts, MLP adaptors) are all cited external components, and there is no load-bearing self-citation chain or imported uniqueness theorem. The principal circularity concern is the evaluation protocol: the central claim 'CoSTeM achieves the best performance' is supported only by 'optimal performances on the validation set,' with the validation set used both for checkpoint/hyperparameter selection and for the final reported numbers. This makes the reported metric a selected optimum rather than a held-out prediction, and the small F1 margin over AIM (1.41 points) is not accompanied by significance tests or repeated runs. The later observation that temporal experts attend to pronucleus, cleavage, and blastocyst stages is a post-hoc interpretation of a model trained on embryologist labels, and while it is framed as confirmation, it does not drive a derived quantitative prediction. Overall, the core architecture is not circularly derived, but the headline quantitative claim is partially fitted to the validation criterion, warranting a moderate score of 4.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard deep learning hyperparameters and assumptions about the clinical dataset. No new physical or theoretical entities are postulated; the spatial and temporal 'experts' are learned parameters, not entities with independent evidence.

free parameters (6)
  • M (number of spatial experts) = 8
    Set by hand in Section 4.1; determines the number of stage-specific feature selectors.
  • K (number of temporal experts) = 8
    Set by hand in Section 4.1; determines the number of learned queries in the temporal selection block.
  • lambda (diversity loss weight) = 0.1
    Set in Section 4.1; balances the cross-entropy and diversity losses.
  • T (sampled frames per video) = 64
    Sampling choice in Section 4.1; frames every 2 hours between 16 and 142 hours post-fertilization.
  • D (temporal transformer depth) = 2
    Architecture choice in Section 3.1; number of transformer blocks in the morphokinetic branch.
  • L (number of temporal selection layers) = 2
    Architecture choice in Section 3.3; number of cross-attention layers in the temporal selection block.
assumptions (5)
  • domain assumption CLIP-pretrained image features transfer to time-lapse embryo microscopy frames.
    The model uses a CLIP ViT-B/16 encoder (Section 3.1) and assumes its features provide a useful representation for embryo morphology without additional domain-specific pretraining.
  • domain assumption The embryologist grading labels used as ground truth are correct and consistent.
    Section 2 states each video was categorized by embryologists through a multi-stage evaluation, but no inter-observer reliability or detailed protocol is given.
  • domain assumption The single-center dataset is representative of the general IVF population.
    Section 2 describes data from one hospital (anonymized as '*** Hospital') using the EmbryoScope system, with no multi-center or external validation.
  • standard math Standard training procedures (cross-entropy loss, AdamW, cosine schedule) converge to a good solution.
    Section 4.1 specifies the optimizer and schedule; this relies on standard optimization assumptions accepted in the deep learning literature.
  • domain assumption Sampling 64 frames at 2-hour intervals retains the morphokinetic events needed for grading, and zero-padding missing frames is harmless.
    Section 4.1 describes the sampling strategy; missing frames are zero-padded, which may introduce artifacts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Time-Lapse Video-Based Embryo Grading via Complementary Spatial-Temporal Pattern Mining." pith.science (2026). https://pith.science/paper/VTQJVEOL

@misc{pith2026250604950,
  author       = {Pith},
  title        = {Pith review of: Time-Lapse Video-Based Embryo Grading via Complementary Spatial-Temporal Pattern Mining},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VTQJVEOL}},
  note         = {Machine review of arXiv:2506.04950}
}
read the original abstract

Artificial intelligence has recently shown promise in automated embryo selection for In-Vitro Fertilization (IVF). However, current approaches either address partial embryo evaluation lacking holistic quality assessment or target clinical outcomes inevitably confounded by extra-embryonic factors, both limiting clinical utility. To bridge this gap, we propose a new task called Video-Based Embryo Grading - the first paradigm that directly utilizes full-length time-lapse monitoring (TLM) videos to predict embryologists' overall quality assessments. To support this task, we curate a real-world clinical dataset comprising over 2,500 TLM videos, each annotated with a grading label indicating the overall quality of embryos. Grounded in clinical decision-making principles, we propose a Complementary Spatial-Temporal Pattern Mining (CoSTeM) framework that conceptually replicates embryologists' evaluation process. The CoSTeM comprises two branches: (1) a morphological branch using a Mixture of Cross-Attentive Experts layer and a Temporal Selection Block to select discriminative local structural features, and (2) a morphokinetic branch employing a Temporal Transformer to model global developmental trajectories, synergistically integrating static and dynamic determinants for grading embryos. Extensive experimental results demonstrate the superiority of our design. This work provides a valuable methodological framework for AI-assisted embryo selection. The dataset and source code will be publicly available upon acceptance.

Figures

Figures reproduced from arXiv: 2506.04950 by the authors.

Figure 1
Figure 1. Representative frames extracted at key developmental stages of a embryo from "good quality" category and robust spatiotemporal reasoning to identify fragmented morphological fea￾tures across various developmental stages for comprehensive quality assessment. As a powerful baseline, we introduce a clinically motivated framework termed CoSTeM that jointly models complementary morphological and morphokinetic patterns fo… view at source ↗
Figure 2
Figure 2. Architecture of the proposed CoSTeM framework. (a) Schematic overview of our framework. (b) Bypass Adaption Network for feature modulation and integration. (c) Mixture of Cross-Attentive Experts (MCAE) layer for adaptive spatial feature selection. (d) Temporal Selection Block (TSB) for local frame selection. training and validation subsets in a 7:3 ratio using stratified sampling to pre￾serve class balance [PITH_FU… view at source ↗
Figure 3
Figure 3. Ablation results of TSB. (a) is the performance comparisons of Temporal Se￾lection Block and average pooling. (b) is the temporal density distribution of selected frames by temporal experts in TSB [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 23 canonical work pages

  1. [1]

    In: Proceedings of the IEEE/CVF International Confer- ence on Computer Vision

    Arnab, A., Dehghani, M., Heigold, G., Sun, C., Lučić, M., Schmid, C.: Vivit: A video vision transformer. In: Proceedings of the IEEE/CVF International Confer- ence on Computer Vision. pp. 6836–6846 (2021)

    work page 2021

  2. [2]

    Reproductive Biomedicine Online22(6), 632–646 (2011)

    Balaban, B., Brison, D., Calderon, G., Catt, J., Conaghan, J., Cowan, L., Ebner, T., Gardner, D., Hardarson, T., Lundin, K., et al.: Istanbul consensus workshop on embryo assessment: proceedings of an expert meeting. Reproductive Biomedicine Online22(6), 632–646 (2011)

    work page 2011

  3. [3]

    Frontiers in Artificial Intelligence7, 1375474 (2024)

    Borna, M.R., Sepehri, M.M., Maleki, B.: An artificial intelligence algorithm to select most viable embryos considering current process in ivf labs. Frontiers in Artificial Intelligence7, 1375474 (2024)

    work page 2024

  4. [4]

    In: European Conference on Computer Vision

    Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., Zagoruyko, S.: End- to-end object detection with transformers. In: European Conference on Computer Vision. pp. 213–229. Springer (2020)

    2020

  5. [5]

    In: International Conference on Medical Image Com- puting and Computer-Assisted Intervention

    Chen, T., Cheng, Y., Wang, J., Yang, Z., Zheng, W., Chen, D.Z., Wu, J.: Au- tomating blastocyst formation and quality prediction in time-lapse imaging with adaptive key frame selection. In: International Conference on Medical Image Com- puting and Computer-Assisted Intervention. pp. 445–455. Springer (2022)

    work page 2022

  6. [6]

    In: 2023 4th International Conference on Electronics and Sustainable Communication Systems (ICESC)

    Eswaran,M.,Balasubramanie,P.,Salman,J.M.:Deeplearningalgorithmsfortime- lapse image sequence-based automated blastocyst quality detection. In: 2023 4th International Conference on Electronics and Sustainable Communication Systems (ICESC). pp. 1246–1252. IEEE (2023)

    work page 2023

  7. [7]

    Journal of Machine Learning Research 23(120), 1–39 (2022)

    Fedus,W.,Zoph,B.,Shazeer,N.:Switchtransformers:Scalingtotrillionparameter models with simple and efficient sparsity. Journal of Machine Learning Research 23(120), 1–39 (2022)

    2022

  8. [8]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Feichtenhofer, C., Fan, H., Malik, J., He, K.: Slowfast networks for video recog- nition. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 6202–6211 (2019)

    work page 2019

Show all 26 references
  1. [9]

    In: 2024 14th International Conference on Cloud Computing, Data Science & Engineering (Confluence)

    Garg, K., Dev, A., Bansal, P., Mittal, H.: An efficient deep learning model for embryo classification. In: 2024 14th International Conference on Cloud Computing, Data Science & Engineering (Confluence). pp. 358–363. IEEE (2024)

    2024

  2. [10]

    Hara, K., Kataoka, H., Satoh, Y.: Can spatiotemporal 3d cnns retrace the history of 2d cnns and imagenet? In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 6546–6555 (2018)

    2018

  3. [11]

    Sun et al

    He, C., Karpaviči¯ ut˙ e, N., Hariharan, R., Jacques, C., Chambost, J., Malmsten, J., Zaninovic, N., Wouters, K., Fréour, T., Hickman, C., et al.: Embryo graphs: 10 Y. Sun et al. Predicting human embryo viability from 3d morphology. In: International Confer- ence on Medical Im...

    2024

  4. [12]

    In: International Conference on Machine Learning

    Houlsby, N., Giurgiu, A., Jastrzebski, S., Morrone, B., De Laroussilhe, Q., Ges- mundo, A., Attariyan, M., Gelly, S.: Parameter-efficient transfer learning for nlp. In: International Conference on Machine Learning. pp. 2790–2799. PMLR (2019)

    2019

  5. [13]

    Human Re- production Update21(4), 411–426 (2015)

    Inhorn, M.C., Patrizio, P.: Infertility around the globe: new thinking on gender, reproductive technologies and global movements in the 21st century. Human Re- production Update21(4), 411–426 (2015)

    2015

  6. [14]

    NPJ Digital Medicine2(1), 21 (2019)

    Khosravi, P., Kazemi, E., Zhan, Q., Malmsten, J.E., Toschi, M., Zisimopoulos, P., Sigaras, A., Lavery, S., Cooper, L.A., Hickman, C., et al.: Deep learning enables robust assessment and selection of human blastocysts after in vitro fertilization. NPJ Digital Medicine2(1), 21 (2019)

    2019

  7. [15]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Kim, J., Shi, Z., Jeong, D., Knittel, J., Yang, H.Y., Song, Y., Li, W., Li, Y., Ben-Yosef, D., Needleman, D., et al.: Multimodal learning for embryo viability prediction in clinical ivf. In: International Conference on Medical Image Computing and Computer-Assisted Intervention...

    2024

  8. [16]

    Computers in Biology and Medicine115, 103494 (2019)

    Kragh, M.F., Rimestad, J., Berntsen, J., Karstoft, H.: Automatic grading of human blastocysts from time-lapse imaging. Computers in Biology and Medicine115, 103494 (2019)

    2019

  9. [17]

    Communications Biology4(1), 415 (2021)

    Liao, Q., Zhang, Q., Feng, X., Huang, H., Xu, H., Tian, B., Liu, J., Yu, Q., Guo, N., Liu, Q., et al.: Development of deep learning algorithms for predicting blastocyst formation and quality by time-lapse monitoring. Communications Biology4(1), 415 (2021)

    2021

  10. [18]

    In: International Conference on Medical Image Computing and Computer Assisted Intervention

    Lockhart, L., Saeedi, P., Au, J., Havelock, J.: Automating embryo development stage detection in time-lapse imaging with synergic loss and temporal learning. In: International Conference on Medical Image Computing and Computer Assisted Intervention. pp. 540–549. Springer (2021)

    2021

  11. [19]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Nguyen, T.P., Pham, T.T., Nguyen, T., Le, H., Nguyen, D., Lam, H., Nguyen, P., Fowler, J., Tran, M.T., Le, N.: Embryosformer: Deformable transformer and collaborative encoding-decoding for embryos stage development classification. In: Proceedings of the IEEE/CVF Winter Confere...

    2023

  12. [20]

    In: International Conference on Cloud Computing, Performance Computing, and Deep Learning (CCPCDL 2023)

    Ou, Z., Zhang, Q., Li, Y., Meng, X., Wang, Y., Ouyang, Y., Li, Z., Zeng, K.: Classi- fication of human embryos by using deep learning. In: International Conference on Cloud Computing, Performance Computing, and Deep Learning (CCPCDL 2023). vol. 12712, pp. 342–347. SPIE (2023)

    2023

  13. [21]

    In: International Conference on Machine Learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning. pp. 8748–8763. PMLR (2021)

    2021

  14. [22]

    Nature Communications15(1), 7756 (2024)

    Rajendran, S., Brendel, M., Barnes, J., Zhan, Q., Malmsten, J.E., Zisimopoulos, P., Sigaras, A., Ofori-Atta, K., Meseguer, M., Miller, K.A., et al.: Automatic ploidy prediction and quality assessment of human blastocysts using time-lapse imaging. Nature Communications15(1), 77...

    2024

  15. [23]

    Truong, V.V., Le, M.H., Phu, P.L., Phan, H., Le, T., Nguyen, H., Pham, V.D., Le, X.H.:Adeeplearningapproachtoembryoqualityassessment.In:TheInternational Conference on Intelligent Systems & Networks. pp. 452–461. Springer (2024)

    2024

  16. [24]

    In: 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)

    Xie, X., Yan, P., Cheng, F.Y., Gao, F., Mai, Q., Li, G.: Early prediction of blasto- cyst development via time-lapse video analysis. In: 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI). pp. 1–5. IEEE (2022) Title Suppressed Due to Excessive Length 11

    2022

  17. [25]

    In: The Eleventh International Conference on Learning Representations (2023)

    Yang, T., Zhu, Y., Xie, Y., Zhang, A., Chen, C., Li, M.: Aim: Adapting image mod- els for efficient video action recognition. In: The Eleventh International Conference on Learning Representations (2023)

    2023

  18. [26]

    In: Pattern Recognition

    Zeman, A., Maerten, A.S., Mengels, A., Sharon, L.F., Spiessens, C., de Beeck, H.O.: Deep learning for human embryo classification at the cleavage stage (day 3). In: Pattern Recognition. ICPR International Workshops and Challenges: Virtual Event, January 10–15, 2021, Proceeding...

    2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.