Table 5.
Oocyte assessment studies using artificial intelligence
| Study | Outcomes of interest | Dataset | AI methods | Results |
|---|---|---|---|---|
| Kanakasabapathy et al. (2020)72* | Whether the addition of synthetic oocyte images generated by a pretrained GAN would improve the performance of a CNN in oocyte assessment. | -Synthetic CNN trained using 1411 oocyte images and 1340 synthetic oocyte images generated by a GAN. | Deep learning—CNN and synthetic GAN. | Synthetic oocyte images generated by a pretrained GAN was able to help a CNN outperform conventionally trained CNN to determine oocytes that fertilized normally or abnormally (67.0 vs 82.6% accuracy). |
| Nayot et al. (2020)74* | CNN based visual assessment tool to predict fertilization and blastocyst development compared to expert embryologists. | CNN based on 17,659 2D oocyte images. Validation studies consisting of balanced 300 oocyte images (100 failed fertilization, 100 fertilized but did not reach blastocyst stage, 100 that reached blastocyst stage). | VIOLET™ (Future Fertility) deep learning AI image analysis tool (CNN). |
-Violet outperformed 17 embryologists from 8 IVF clinics to accurately predict fertilization (71.7% vs 58.9%) and blastocyst development (62.8% vs 52.2%). -Reproducible results in a second validation study. -AI outperforms manual assessment in oocyte morphology assessment. |
| Mercuri et al. (2022)75* | Oocyte images analyzed and scored by image analysis AI tool predicting quality of blastocyst development. | 16261 oocyte images from 5620 subjects with known clinical outcomes | MAGENTA™ (Future Fertility) AI image analysis tool. |
-Magenta tool score correlated with blastocyst quality in stepwise manner. -Tool was able to differentiate between non-blastocyst and low quality blastocyst (ICM or TE grade of C or D) as well as low quality blastocyst and medium/high quality blastocyst (ICM and TE grade of A or B). |
| Link et al. (2022)76* | Prediction of oocyte developmental potential to top quality day-5 blastocyst from cumulus oophorus cells compared to expert embryologist. | 65 cumulus cell samples from oocytes of 26 patients. | 8 ML models and 25-gene network—OsteraTest bioinformatics tool. |
-Cumulus cells from oocytes underwent real-time PCR with 25 target genes. Gene expression levels are computed by ML models to indicate developmental potential of each oocyte. -86% accuracy in predicting oocyte developmental capacity into a top quality blastocyst. -Yet to undergo a large-scale, prospective, randomized study for external validation. |
Summary of studies using artificial intelligence (AI) and machine learning (ML) methods for oocyte assessment, prediction, and selection. The asterisk (*) indicates studies from conference proceedings. CNN convolutional neural network, GAN generative adversarial network, ICM inner cell mass, TE trophectoderm, PCR polymerase chain reaction, AUC area under curve.