A generalized AI system for human embryo selection covering the entire IVF cycle via multi-modal contrastive learning
Creators
- 1. State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications
- 2. College of Future Technology, Peking University and Peking-Tsinghua Center for Life Sciences
- 3. Institute of Reproductive and Stem Cells, School of Basic Medicine, Central South University
- 4. Reproductive Medicine Center, the First Affiliated Hospital, Sun Yat-sen University
- 5. Research Department, CITIC Xiangya Reproductive and Genetic Hospital
- 6. Capital Institute of Pediatrics, Affiliated Children's Hospital
- 7. The First College of Clinical Medical Science, China Three Gorges University & Yichang Central People's Hospital
- 8. UCL Cancer Institute, University College London
Description
This is the pytorch implementation of our paper "A generalized AI system for human embryo selection covering the entire IVF cycle via multi-modal contrastive learning". In this paper, we advocate bringing a wealth of 2D embryo images as compensation for the lack of 3D time-lapse data, aiming to build a universal self-supervised representation learning framework for embryo assessments. We propose a novel self-supervised learning framework, named VTCLR (Visual-Temporal Contrastive Learning of Representations) to learn multimodal embryo representations from temporal videos and static images with pretraining on large unlabeled data, with a Transformer-based network backbone, IVFormer (Image Video TransFormer). We conduct expensive experiments on 3D/2D embryo image analysis tasks, including morphology assessments, live birth prediction and PGT prediction. The results show that the proposed VTCLR and IVFormer achieves promising performance on various downstream tasks, outperforming the ImageNet pre-training and other advanced SSL counterparts substantially.
Files
blastocyst.json
Files
(461.3 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:ad9b5a6eb5fe189cbefc2a061eafb332
|
48.4 kB | Download |
|
md5:f8bc735f6f26f1b5d46160483e4bc4cc
|
626 Bytes | Preview Download |
|
md5:7fa92a49eeca3add57d771b6c7c9275e
|
645 Bytes | Preview Download |
|
md5:17e291b85efd00ace83dec61f8c091dd
|
2.5 kB | Download |
|
md5:cab3bc9a2fdf2997eca69e071d25c0e1
|
1.3 kB | Download |
|
md5:a1d12c479545762d8d96885a1847d797
|
15.1 kB | Download |
|
md5:c32844955fd7c6758487ead996e6e7a1
|
9.8 kB | Download |
|
md5:cc090f5a8ba31c051e73116ef33b5780
|
5.5 kB | Download |
|
md5:59f1c5dd7b58b0d9969d4afe8a487260
|
328.3 kB | Preview Download |
|
md5:d192227e698f7d2d742ac2e35eda4b6e
|
3.1 kB | Download |
|
md5:c32844955fd7c6758487ead996e6e7a1
|
9.8 kB | Download |
|
md5:a15cf0ae661fba3b533b4060ffc98ef4
|
1.1 kB | Download |
|
md5:64d4ae42928a51d4cf7d702a5b861bcc
|
612 Bytes | Preview Download |
|
md5:28e79b8572f45c2f41b72425725a8248
|
3.7 kB | Download |
|
md5:c348a1df5a984a45aaec567c090f4999
|
16.1 kB | Download |
|
md5:48bbc127da757e50fc74d7216e2818ca
|
593 Bytes | Preview Download |
|
md5:226b52529a3ff12d8407329427332881
|
554 Bytes | Preview Download |
|
md5:868f6320e011b69cad9688509fa75331
|
3.1 kB | Download |
|
md5:da5c5e94074b73205675e94c82967fac
|
634 Bytes | Preview Download |
|
md5:84f5142f0655fb3b6e2700ec7fd4f57a
|
3.0 kB | Preview Download |
|
md5:281c35dd412770b2ff668af9b92d4e8e
|
6.9 kB | Download |
