Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

EmbryoSorter: An explainable deep learning model for distinguishing clinical and biochemical pregnancy outcomes in IVF

Liu, Zebang, Ai, Jingqi, Yu, Xinning, Kang, Shuping, Guo, Zijun, Huang, Chuyu, Li, Huixi and Liu, Jing 2026. EmbryoSorter: An explainable deep learning model for distinguishing clinical and biochemical pregnancy outcomes in IVF. Procedia Computer Science 292 , pp. 5721-5730. 10.1016/j.procs.2026.10.143

[thumbnail of 1-s2.0-S1877050926038044-main.pdf] PDF - Published Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (886kB)

Abstract

In vitro fertilization (IVF) is widely used to treat infertility, yet accurately distinguishing clinical pregnancy from biochemical pregnancy remains challenging. Although artificial intelligence has been applied to embryo assessment, existing models have not specifically addressed this distinction using static embryo images or supported both single and double embryo transfers. We propose EmbryoSorter, a deep learning framework designed to fill this gap. Using a retrospective dataset of 1,822 embryo images from 1,215 patients, we developed a model based on a DenseNet169 backbone with a Feature Pyramid Network for multi-scale feature extraction and a novel Trophectoderm–Inner Cell mass Attention (TEIC-Attn) module to adaptively fuse features in double-embryo transfer cases. The architecture supports both single and double transfers and provides explainable weighting of individual embryos. EmbryoSorter achieved an AUC of 0.8415, accuracy of 79.01%, sensitivity of 97.56%, and specificity of 60.00%, outperforming benchmark CNN models. Gradient-weighted Class Activation Mapping (Grad-CAM) analysis demonstrated that predictions focused on biologically relevant regions, including the inner cell mass and trophectoderm. This study presents a clinically applicable and explainable framework for distinguishing early pregnancy outcomes from static embryo images, supporting improved embryo selection in IVF practice.

Item Type: Article
Date Type: Published Online
Status: Published
Schools: Schools > Engineering
Publisher: Elsevier
ISSN: 1877-0509
Date of First Compliant Deposit: 9 November 2026
Last Modified: 09 Nov 2026 11:45
URI: https://orca.cardiff.ac.uk/id/eprint/190758

Actions (repository staff only)

Edit Item Edit Item

Downloads

Downloads per month over past year

View more statistics