P–171 Automated oocyte and zygote denudation using a novel microfluidic device supervised by a computer vision algorithm

Abstract

Abstract Study question Is it possible to remove cumulus cells using a 16-well microfluidic device with automated flows to facilitate vitrification, ICSI, NI-PGT or non-invasive metabolomics analysis? Summary answer The designed automated system and protocol efficiently denude 16 samples simultaneously with a x10 lower shear stress than the manual process and without human intervention. What is known already Most processes involved in IVF such as insemination, washing, denudation, embryo culture and selection are still manually performed, labor-intensive and require highly skilled professionals. This leads to a significant variability in the clinical outcomes achieved by different embryologists and labs. The automation of these processes is a promising approach to reduce costs and improve the accessibility to assisted reproductive therapies. Although a simple procedure, standardization of cumulus oocyte complex (COCs) and zygotes denudation is key to facilitate ICSI, vitrification and to avoid DNA contamination for NI-embryo testing (PGT or metabolomics), while avoiding damage to the oocyte by excessive shear stress. Study design, size, duration A total of 160 cow COCs were used due to their size similarity with human COCs. Half were denuded 16–20 hours post-insemination and half pre-insemination for 5–10 minutes. COCs were classified as partially denuded if fertilization assessment, ICSI or vitrification was possible, and completely denuded if no cumulus cells remained. COCs controls were manually denuded (Stripper® capillary 145μm ID) to compare shear stress between procedures. This study was conducted during 2020 – 2021. Participants/materials, setting, methods We developed a customized microfluidic biochip that exerts a particular fluid motion while avoiding egg entrapment within microfluidic channels. The denudation efficacy was established by subjectively scoring images of bovine oocytes after generating a continuous “Push & Pull” fluid motion inside the biochip wells. A Computer Vision model was developed in parallel in order to optically assess denudation completion. The model used was a Pytorch implementation of Faster-RCNN with ImageNet pretrained weights Main results and the role of chance 96 bovine COCs were microfuidically handled post insemination achieving complete (56/96) or partial (40/96) removal of the cumulus cells on day 1, while for day 3 double denudation group, 89/96 (92.7%) were completely denuded while the rest remained partially denuded. In comparison, 80/80 (100%) of manually denuded cow COCs, achieved complete denudation (50% post-insemination group and 50% pre-insemination group). In addition, 48/64 (75%) cow COCs treated pre-insemination were partially denuded, enough to carry out ICSI after 5–10 min of treatment. The results here obtained indicate that media needs to flow through the device at a rate that can generate enough shear to strip off the cumulus-corona cells while avoiding emptying of the reservoirs containing the fertilization or culture medium. The shear stress of our design was calculated to be smaller than 4.4 Pa, about ten times lower than the one applied by the manual process (∼44Pa). The deep learning algorithm was tested on 20 unseen human oocytes on day 1, with 10 true positives 9 true negatives, and 1 false negative (95% accuracy). Limitations, reasons for caution The success of the denudation procedure was dependent on the design of the biochip wells and the microfluidic protocol used. The accuracy of our findings is still limited because of the difficulty in manufacturing prototype biochips. Wider implications of the findings: Complete denudation is key to avoid DNA contamination for NI-PGT or metabolomics analysis, while avoiding damage to the oocyte by excessive shear stress. Our device, which has the potential of scaling up and treat each oocyte individually, can improve automation and increase efficiency of current ART procedures Trial registration number NA