Sperm Selection via Machine Learning Morphology Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current sperm analysis techniques, such as CASA and Raman spectroscopy, are inaccurate in predicting sperm morphology and DNA integrity, leading to low fertilization and embryo development rates, and often render sperm unusable due to damage from fixation techniques.
Innovation Solution
A machine learning-trained specimen analysis model uses vibrational microspectroscopy and quantitative phase imaging to generate specimen scores for individual sperm without damaging them, allowing for real-time selection based on imaging data that indicates successful fertilization and embryo development potential.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If CASA systems are used to automate sperm analysis, then analysis speed and throughput are improved, but measurement precision of individual sperm morphology deteriorates
Solution Approach 1:
The system segments the sperm analysis process into two distinct stages: (1) CASA-based automated screening to rapidly identify motile sperm candidates, and (2) deep learning-based individual sperm morphology assessment to precisely evaluate selected specimens. This segmentation allows high-throughput initial screening while maintaining high precision in the final selection decision.
Solution Approach 2:
The patent introduces an intermediary deep learning model that acts as a bridge between CASA automated analysis and final sperm selection. The model takes CASA output data as input and performs detailed morphology assessment, serving as an intermediary layer that combines the speed of automated systems with the precision needed for individual sperm evaluation.
2Measurement precision
If fixation techniques are used in Raman spectroscopy, then measurement precision of sperm quality is improved, but sperm integrity deteriorates
Solution Approach 1:
The patent extracts and removes the harmful fixation step from the traditional Raman spectroscopy workflow. Instead of using fixation to improve measurement precision, the system uses alternative deep learning-based image analysis methods that achieve high precision without causing sperm damage, thereby extracting the harmful element while preserving sperm integrity.
Solution Approach 2:
The patent substitutes the mechanical/chemical fixation process with a computational deep learning analysis system. Instead of physically fixing sperm samples to improve measurement quality, the system uses advanced image processing and neural networks to achieve high-precision morphology assessment without physical or chemical intervention that could damage the sperm.
3Measurement precision
If manual microscopy assessment is used for sperm selection, then measurement precision of morphology is improved, but productivity deteriorates
Solution Approach 1:
The system segments the sperm selection workflow into automated CASA pre-screening and manual/deep learning-based final assessment. This allows the majority of sperm to be rapidly filtered through automated systems while expert-level morphology assessment is applied only to selected candidates, combining high throughput with high precision.
Solution Approach 2:
The patent uses deep learning models trained on extensive datasets to create computational copies of expert embryologist decision-making. These models replicate human expertise in morphology assessment, enabling automated systems to achieve measurement precision comparable to manual expert review while dramatically increasing productivity.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The model provides improved fertilization and embryo development outcomes by accurately differentiating between viable and non-viable sperm, maintaining sperm integrity, and enabling real-time selection for assisted reproduction technologies.
Implementation Method 1
In various embodiments, the specimen imaging data corresponds to vibrational spectra of the one or more specimen captured using vibrational microspectroscopy
Implementation Method 2
In various embodiments, the specimen imaging data corresponds to phase imaging data of the one or more specimen captured using quantitative phase imaging
Data Source
AI summary
A user computing entity provides a specimen score for use in selecting a sperm for use in a fertilization event. The user computing entity provides the score by obtaining specimen image data comprising imaging data of one or more specimen, the imaging data corresponds to at least one type of imaging; generating a specimen scoring request including the specimen image data; providing the specimen scoring request for receipt by a network computing entity; receiving a specimen response comprising a respective specimen score for the one or more specimen, the respective specimen score for the one or more specimen generated by a machine-learning trained specimen analysis model; processing the respective specimen score for the one or more specimen to generate a graphical representation of the respective specimen score; and causing display of the graphical representation of the respective specimen score.