WISE, our advanced self-supervised learning encoder designed to enhance embryo identification in IVF across time-lapse, cross-device, and cryo-thawed scenarios. Based on Vision Transformer (ViT) architecture and masked autoencoders (MAE), WISE was trained using three datasets: SSL pre-training, time-lapse identification, and cross-device identification.

WISE processes embryo images through object detection, cropping, padding, and resizing, achieving impressive accuracy rates: 99.89% for time-lapse, 83.55% for cross-device, and 82.22% for cryo-thawed scenarios. Our method showed a 10% improvement in cross-device and cryo-thawed identification tasks compared to traditional methods. Additionally, WISE outperformed embryologists by 9.5%, 12%, and 18% in these scenarios.

This study is the first to apply SSL in embryo identification, demonstrating WISE's potential for future applications in embryo witnessing, significantly improving accuracy and reliability.
Features & Benefits

WISE Demonstrated Competence

Our study showcases WISE's proficiency in time-lapse, cross-device, and cryo-thawed tasks

Enhanced Accuracy with SSL Method

Leveraging SSL method, MAE, enhances accuracy, highlighting the efficacy of these techniques in augmenting models with limited labeled data

Comparing Embryologists

Compared to the performance of embryologists, WISE demonstrates remarkable potential in real-world applications

Pioneering Solution with ViT and MAE

This study pioneers the integration of ViT and MAE in IVF, offering a cutting-edge solution for embryo identification and reducing non-conformance in laboratory and IVF treatment processes