System and method for automated gamete selection

Abstract

In variants, a method for automated gamete selection can include: sampling a video of a scene having a plurality of gametes, tracking each gamete across successive images, and determining attribute values for a gamete, and selecting the gamete. The attribute values can be determined using a model trained to predict the attribute values for the gamete based on a video.

We claim: 
     
         1 . A method, comprising:
 sampling a video of a gamete; and   automatically determining a DNA fragmentation index (DFI) score for the gamete based on the video, using a trained model;   wherein the gamete is selected from a set of gametes based the DFI score.   
     
     
         2 . The method of  claim 1 , wherein the trained model comprises a machine learning model trained by:
 sampling a training video of a training gamete;   determining a DFI score for the training gamete based on a DFI measurement for the training gamete; and   training a model to predict the DFI score for the training gamete based on the training video.   
     
     
         3 . The method of  claim 1 , further comprising automatically determining a confidence parameter based on the video, wherein the gamete is further selected based the confidence parameter. 
     
     
         4 . The method of  claim 1 , wherein the DFI score is one of a set of DFI scores determined based on the video using the trained model, wherein each DFI score in the set corresponds to a different time window, wherein the gamete is selected based on the set of DFI scores. 
     
     
         5 . The method of  claim 4 , further comprising determining a confidence parameter based on the set of DFI scores, wherein the gamete is further selected based the confidence parameter. 
     
     
         6 . The method of  claim 4 , further comprising determining a distribution of the DFI scores, wherein the gamete is selected from the set of gametes based on the distribution of DFI scores. 
     
     
         7 . The method of  claim 1 , further comprising automatically determining an attribute value for the gamete based on the video using a second trained model, wherein the gamete is further selected based the attribute value. 
     
     
         8 . The method of  claim 7 , wherein the attribute value comprises at least one of a selection score, a motility score, or a morphology score. 
     
     
         9 . The method of  claim 7 , wherein the second trained model is a machine learning model, trained by:
 providing a video of a training gamete to a set of specialists;   from each specialist in the set, receiving a selection metric for the training gamete;   determining an overall metric for the training gamete based on the selection metrics; and   training a second model to predict the overall metric for the training gamete based on the video of the training gamete.   
     
     
         10 . The method of  claim 9 , wherein the overall metric comprises a selection probability. 
     
     
         11 . The method of  claim 7 , wherein the gamete is selected based on the DFI score and the attribute value using a set of heuristics. 
     
     
         12 . The method of  claim 7 , wherein the gamete is selected based on the DFI score and the attribute value using a selection model, wherein the selection model is trained using development data collected for an embryo fertilized using a training gamete. 
     
     
         13 . The method of  claim 1 , wherein the DFI score is determined without staining the gamete. 
     
     
         14 . The method of  claim 1 , wherein the gamete is selected for an assistive reproductive technology. 
     
     
         15 . A method, comprising:
 sampling a video of a gamete; and   automatically determining a destructive attribute value for the gamete based on the video, using a model, wherein the model is trained by:
 sampling a training video of a training gamete; 
 determining a destructive attribute value for the training gamete based on a measurement for the training gamete obtained using a destructive assay; and 
 training the model to predict the destructive attribute value for the training gamete based on the training video; 
   wherein the gamete is selected based the destructive attribute value.   
     
     
         16 . The method of  claim 15 , wherein the measurement obtained using the destructive assay comprises a measurement for at least one of: DNA fragmentation index (DFI), DNA condensation level, biochemical marker analysis, or vitality. 
     
     
         17 . The method of  claim 15 , wherein the destructive assay comprises at least one of: acridine orange test (AO), sperm chromatin structure assay (SCSA), deoxynucleotidyl transferase-mediated dUTP nick end labeling assay (TUNEL), single-cell gel electrophoresis assay (COMET), or sperm chromatin dispersion test (SCD). 
     
     
         18 . The method of  claim 15 , wherein the trained model comprises an attention mechanism configured to focus on frames depicting a flat side of the gamete. 
     
     
         19 . The method of  claim 15 , further comprising extracting non-semantic features for the gamete based on the video, wherein the DFI score for the gamete is determined based on the extracted non-semantic features. 
     
     
         20 . The method of  claim 15 , wherein the destructive attribute value for the gamete is determined without destroying the gamete.