Real-Time Automated Detection of Human Cumulus-Oocyte Complexes Using High-Resolution Imaging and AI Inference

Mendizabal-Ruiz et al., American Society for Reproductive Medicine (ASRM) 2025 Scientific Congress & Expo


Objective

To develop a fully automated system capable of detecting human cumulus-oocyte complexes (COCs) from follicular fluid during egg retrieval, combining high-resolution imaging with real-time artificial intelligence.

Materials and Methods

We designed a novel system integrating a 20-megapixel digital camera and a custom-built collimated backlight illumination stage. Upon detection of a Petri dish placement, the system automatically captures a high-resolution image (5488×3672 pixels) and perform a real-time AI inference. The detection model was based on RT-DETR, a transformer-based neural network architecture, trained on 640×640 slices with Slicing-Aided Hyper Inference (SAHI). A total of 78 original images (149 annotated COCs) were collected, divided into training, validation, and testing sets. Final deployment was achieved through an integrated inference API.

Results

In the test set comprising 30 COCs, the system correctly detected 28 COCs (True Positives) and missed 2 COCs (False Negatives), achieving a True Positive Rate (TPR) of 93%. Detection occurred immediately after dish placement without manual input.

Conclusions

This work introduces a real-time, hands-free AI-driven detection platform for cumulus-oocyte complexes.

Impact Statement

The combination of high-resolution imaging and automated inference has the potential to revolutionize oocyte identification workflows, standardizing and accelerating decision-making in IVF laboratories.

Support

This work was supported by Conceivable Life Sciences.

 
Previous
Previous

Pre-Clinical Validation of an AI-Powered Automated Robotic Intracytoplasmic Sperm Injection System

Next
Next

Uncertainty-Aware Shared Control for Vision-Based Micromanipulation