Training a machine learning model to assess embryo characteristics from video image data
A method for training a computer-based machine learning model to assess a characteristics-of-interest in embryos such as viability, sex or genetic superiority/inferiority by processing video of each of a plurality of embryos and wherein a subset of the embryos each embodies the characteristic-of-interest. The method includes obtaining (taking or otherwise procuring) a plurality of training videos, each video being of a target embryo and each video having a real-time frame speed (specifically not time-lapse frame speed) over a continuous recording duration of ten minutes or less. Each video includes image data representing micro-movement of the target embryo. The image data is processed thereby generating a computer-based machine learning model for assessing the characteristic-of-interest in future embryos from image data of those embryos.
1 . A method for training a computer-based machine learning model to assess an embryo characteristic-of-interest by processing video of each of a plurality of embryos and wherein a subset of said embryos each embodies the characteristic-of-interest, the method comprising:
obtaining a plurality of training videos, each video being of a target embryo and each video having a real-time frame speed over a continuous recording duration of ten minutes or less and wherein each video comprises image data representing micro-movement of the target embryo;
pairing said image data to known outcome data of the characteristic-of-interest; and
analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning to thereby generate a computer-based machine learning model.
2 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that established a pregnancy.
3 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that produced a livebirth offspring.
4 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that established a specific-sex pregnancy.
5 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that produced a specific-sex livebirth offspring.
6 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that established a male-sex pregnancy.
7 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that produced a male-sex livebirth offspring.
8 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that established a female-sex pregnancy.
9 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that produced a female-sex livebirth offspring.
10 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing embryo outcome data that represents embryo transfers into recipients that produced a genetically superior offspring.
11 . The method of claim 1 , wherein analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning includes analyzing correlated embryo outcome data that represents embryo transfers into recipients that produced a genetically inferior offspring.
12 . The method of claim 1 , wherein said video has a continuous recording duration of thirty seconds or less.
13 . The method of claim 1 , wherein said video has a continuous recording duration greater than thirty seconds.
14 . The method of claim 1 , wherein said video has a frame speed at least as fast as ten frames per second.
15 . The method of claim 1 , wherein said video has a frame speed at least as fast as two frames per second.
16 . The method of claim 1 , wherein micro-movement of the target embryo comprises morphokinetic movement of the target embryo.
17 . The method of claim 1 , wherein said real-time frame speed video is time-lapse free video.
18 . The method of claim 1 , wherein said real-time frame speed video is prerecorded at a time before the image data is processed.
19 . A system comprising:
one or more processors; and
a computer-readable medium comprising instructions stored therein, which when executed by the one or more processors, cause the one or more processors to:
execute training of a computer-based machine learning model that assesses an embryo characteristic-of-interest by processing video of each of a plurality of embryos and wherein a subset of said embryos each embodies the characteristic-of-interest, said training comprising:
obtaining a plurality of training videos, each video being of a target embryo and each video having a real-time frame speed over a continuous recording duration of ten minutes or less and wherein each video comprises image data representing micro-movement of the target embryo;
pairing said image data to known outcome data of the characteristic-of-interest; and
analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning to thereby generate a computer-based machine learning model.
20 . A non-transitory computer-readable storage medium comprising computer-readable instructions, which when executed by a computing system, execute training of a computer-based machine learning model that assesses an embryo characteristic-of-interest by processing video of each of a plurality of embryos and wherein a subset of said embryos each embodies the characteristic-of-interest, said training comprising:
obtaining a plurality of training videos, each video being of a target embryo and each video having a real-time frame speed over a continuous recording duration of ten minutes or less and wherein each video comprises image data representing micro-movement of the target embryo;
pairing said image data to known outcome data of the characteristic-of-interest; and
analyzing the paired data by determining characteristics in said image data that are indicative of the characteristic-of-interest using machine learning to thereby generate a computer-based machine learning model.