Automated Embryo Viability Analysis with Morphokinetic Feature Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for selecting embryos for in-vitro fertilization (IVF) are subjective, time-consuming, and lack full automation, particularly in evaluating embryo viability and quality, as they rely on manual analysis and are not interpretable by clinicians, and do not account for key features relevant to embryo quality.
Innovation Solution
Employ deep learning techniques, specifically convolutional neural networks (CNNs), to automate the measurement of morphokinetic embryo features, including segmentation, fragmentation detection, developmental stage classification, and pronuclei identification, using a pipeline of fully convolutional networks (FCN), regression CNNs, classification CNNs, and Mask R-CNNs, to provide objective viability scores and rankings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis of embryo morphology is used, then clinicians can evaluate embryo quality, but the process is time-consuming and subjective
Solution Approach 1:
The patent replaces the manual mechanical inspection process with an automated image processing system using convolutional neural networks (CNNs). The system captures embryo images and automatically analyzes morphological features, cell divisions, and developmental stages through deep learning algorithms, eliminating the need for time-consuming manual microscopic examination while maintaining or improving evaluation accuracy.
Solution Approach 2:
The system creates digital copies of embryos through high-resolution imaging and uses these images for analysis instead of directly examining physical embryos under a microscope. This allows multiple analyses to be performed on the same digital copy without disturbing the actual embryo, enabling both rapid automated assessment and potential review by clinicians.
2Productivity
If automated image processing is used, then analysis time is reduced, but interpretability for clinicians is lost
Solution Approach 1:
The system provides feedback to clinicians by presenting automated analysis results in an interpretable format that includes confidence scores, detected morphological features, and developmental stage classifications. This allows clinicians to review and understand the automated system's assessments, maintaining clinical oversight and interpretability while benefiting from automated speed.
Solution Approach 2:
The patent introduces an intermediary layer between automated image processing and clinical decision-making. The CNN system processes images and generates structured outputs that are then presented to clinicians in a comprehensible format, serving as a bridge that maintains both automation efficiency and clinical interpretability.
3Extent of automation
If deep learning models are trained to predict embryo quality directly, then automation is achieved, but the models learn confounding variables instead of true quality indicators
Solution Approach 1:
The patent extracts and removes confounding variables from the training process by carefully curating training data to focus only on morphological features and developmental characteristics that are true indicators of embryo quality. The system excludes data that might introduce bias from external factors such as patient demographics or laboratory conditions, ensuring the model learns genuine quality indicators rather than spurious correlations.
Solution Approach 2:
The system changes the parameters used for training by focusing specifically on morphological parameters and developmental stage parameters that are known to be reliable indicators of embryo quality. By adjusting the training parameters to emphasize these specific features and exclude confounding variables, the model achieves both automation and reliability.
Data Source
AI summary
Embodiments described herein use deep learning to automate measurement of key morphokinetic embryo features associated with viability and quality, in particular those relevant for clinical in-vitro fertilization (IVF). Systems and methods may, for example, acquire one or more digital images of one or more embryos; select one or more embryos in each digital image; and for each selected embryo, (i) computationally characterize the zona pellucida, detect the degree of fragmentation in the embryo, and for each embryo with a low fragmentation score, computationally classify the embryo's developmental stage based on whether cells constituting the embryo exceed a threshold number (e.g., nine). For an embryo consisting of a single cell, pronuclei may be detected and counted. Based on these measurements, a viability score may be assigned to the embryo.