EmbryoProfiler: A Visual Clinical Decision Support System for IVF

Area of Science:

• Reproductive medicine
• Medical imaging analysis
• Computational biology

Background:

• In-vitro fertilization (IVF) is a common treatment for infertility, yet success rates remain low at approximately 20% per cycle.
• Embryo selection is crucial but challenging, with current automated methods lacking transparency and requiring extensive manual input.
• Time-lapse microscopy offers data for computational analysis, but clinical validation of automated tools is difficult.

Purpose of the Study:

• To develop EmbryoProfiler, a visual analytics system for assessing embryo viability using time-lapse microscopy imagery.
• To support clinicians in making more informed embryo selection decisions for improved IVF outcomes.
• To integrate explainable AI for transparent and clinically relevant embryo scoring.

Main Methods:

• Collaborative development of a visual analytics system with embryologists, biologists, and machine learning experts.
• Implementation of a deep learning pipeline for automated image annotation and extraction of clinically interpretable features.
• Design of a semi-automatic, visualization-based workflow with interactive tools like cell-shape plots for embryo assessment.

Main Results:

• The system provides a workflow for fertilization assessment, developmental timing, and morphological inspection.
• Interactive visualizations facilitate efficient analysis of embryo characteristics.
• An explainable machine learning classifier offers transparent embryo viability scoring to predict live birth outcomes.

Conclusions:

• EmbryoProfiler enables clinicians to make better-informed embryo selection decisions.
• The system's visual and explainable AI approach addresses limitations of existing automated methods.
• This tool has the potential to improve clinical outcomes in IVF treatments.