Disclaimer: EMA by AIVF has CE mark. In the US, an AI component of EMA is considered an investigational device. Limited by Federal (or United States) law to investigational use.
©All rights reserved AIVF 2023
Single knowledge center top

Early ploidy markers detected by machine learning applied to computer vision: is it possible to identify aneuploid embryos during the first 48 hours of development?

L.Bori, M. Meseguer, Gimenez C - IVI RMA Valencia, IVF Laboratory, Valencia, Spain; R. Maor, C. Curchoe, D. Seidman, D. Gilboa - AIVF, Tel Aviv, Israel | Nov 21, 2023

OBJECTIVE: To identify early markers of aneuploidy that can be detected by computer vision and artificial intelligence (AI) to predict embryo ploidy among viable blastocysts for trophectodermal biopsy.

MATERIALS AND METHODS: In this single-center study, we used a retrospective dataset consisting of time-lapse images from 555 preimplantation embryos with known ploidy status (394 aneuploid and 261 euploid blastocysts). Images obtained from Embryoscope (Vitrolife, Denmark) were used to measure three morphological features and several mathematical models were built to compare the characteristics between euploid and aneuploid blastocysts with computer vision and machine learning. Firstly, we measured the average difference in grayscale values (lightness vs darkness) between consecutive frames as an estimate for embryo activity. Secondly, we measured the changes in embryo perimeter length between consecutive frames as an estimate for embryo perimeter maintenance.  Finally, we took the measurements of the total area in a single cross-section of the embryo between two frames (insemination and 48 hours post insemination) as a growth estimate.

RESULTS: Out of the possible grayscale values (0-255) between consecutive frames; euploid and aneuploid embryos displayed a similar mean variation: 2.26±0.86 vs. 2.42±1.57 respectively. However, the detailed calculation of the standard deviation (SD) showed that 36 aneuploid embryos had a SD >4; while no embryos had such a deviation in the euploid group. This means that we would have avoided biopsying 9.1% of aneuploid embryos in the first 24 hours of development, without rejecting any euploid embryos. The average difference between consecutive frames for the perimeter length was the same for euploid (1.42±0.99 μm) and aneuploid embryos (1.42±1.08 μm). However, there were 20 aneuploid embryos with a variation of more than 3.4 μm; while only 5 embryos showed such variation in the euploid group. In other words, a threshold set at 3.4μm (36 hours post insemination) could prevent biopsy of 5.1% of aneuploid embryos. Regarding the visible cross-section area, euploid and aneuploid embryos also did not show significantly different growth, in terms of mean: 317±21 μm2 vs. 306±24 μm2 respectively. Nevertheless, 44 aneuploid embryos were unable to grow more than 275 μm2; while only 11 euploid embryos failed to exceed that size. That is, it would be possible to avoid biopsy of 11.2% aneuploid embryos with a threshold of 275 μm2 at 48 hours after insemination.

CONCLUSIONS: Computer vision analysis of grayscale values, embryo perimeter and cross-sectional area in the first 48 hours after insemination would allow discriminating a relevant proportion of blastocysts suitable for biopsy.

IMPACT STATEMENT: The “explainability” and implementation of our AI model allows for a more objective assessment of embryo quality and improves the clinics’ ability to prioritize embryos for preimplantation genetic diagnosis; enhancing preferential transfer with a validated and trusted framework that drastically reduces the chances of transferring an aneuploid embryo to our patients.

Study funding/competing interest(s): This work was supported by the Ministry of Science and Universities CDTI (IDI-20191102) awarded to M.M