Cytoplasmic Strings Analysis in Human Embryo Time-Lapse Videos using Deep Learning Framework

arXiv ID: 2512.09461v1

Published: 2025-12-10

Authors: Anabia Sohail, Mohamad Alansari, Ahmed Abughali, Asmaa Chehab, Abdelfatah Ahmed, Divya Velayudhan, Sajid Javed, Hasan Al Marzouqi, Ameena Saad Al-Sumaiti, Junaid Kashir, Naoufel Werghi

Categories: cs.CV, cs.AI

Relevance Score: 1.00 / 1.00

View on arXiv Download PDF

Summary

This paper introduces the first deep learning framework for automated detection and localization of Cytoplasmic Strings (CS) in human IVF embryo time-lapse videos, addressing the current subjective and labor-intensive manual assessment. By developing a novel Uncertainty-aware Contractive Embedding (NUCE) loss for highly imbalanced data and employing an RF-DETR-based localizer, the framework aims to improve embryo selection by objectively identifying this emerging biomarker.

Medical Relevance

This research directly addresses a critical challenge in reproductive medicine by providing an objective, automated method for assessing a key biomarker for embryo viability. This can significantly improve the efficacy of in-vitro fertilization (IVF) by enhancing embryo selection and ultimately increasing success rates for individuals undergoing infertility treatment.

AI Health Application

The AI application is a two-stage deep learning framework designed to automatically detect and localize 'Cytoplasmic Strings' in human embryo time-lapse videos. This system provides an objective and automated method for assessing a key biomarker associated with embryo viability and quality. By automating this assessment, the AI aims to reduce subjectivity, improve the efficiency and accuracy of embryo selection in IVF procedures, thereby enhancing treatment outcomes for patients undergoing infertility treatment.

Key Points

Infertility is a major global health issue, with embryo selection in in-vitro fertilization (IVF) serving as a critical bottleneck due to reliance on conventional morphokinetic features.
Cytoplasmic Strings (CS) are identified as emerging biomarkers associated with faster blastocyst formation, higher blastocyst grades, and improved viability, yet their assessment is currently manual, subjective, and prone to detection challenges.
A biologically validated CS dataset was curated using a human-in-the-loop annotation pipeline, comprising 13,568 frames from TLI videos with highly sparse CS-positive instances.
A novel two-stage deep learning framework is proposed: first, frame-level classification of CS presence, and second, localization of CS regions in identified positive cases.
The Novel Uncertainty-aware Contractive Embedding (NUCE) loss was introduced to address severe class imbalance and feature uncertainty, coupling confidence-aware reweighting with an embedding contraction term to form compact, well-separated class clusters.
NUCE consistently improved F1-score across five different transformer backbones, demonstrating its effectiveness in challenging imbalanced datasets.
RF-DETR-based localization achieved state-of-the-art (SOTA) detection performance specifically for thin, low-contrast CS structures, which are inherently difficult to identify.

Methodology

The study utilized a human-in-the-loop annotation pipeline to create a biologically validated dataset of 13,568 human IVF embryo time-lapse video frames, which contained highly sparse CS-positive instances. A two-stage deep learning framework was then developed: the first stage performs frame-level classification to determine CS presence, and the second stage localizes the CS regions within positive frames using an RF-DETR-based model. To mitigate challenges posed by severe class imbalance and feature uncertainty, a novel Novel Uncertainty-aware Contractive Embedding (NUCE) loss function was introduced, which combines confidence-aware reweighting with an embedding contraction term.

Key Findings

The research successfully developed the first computational framework for automated Cytoplasmic Strings (CS) analysis in human IVF embryos. The proposed Novel Uncertainty-aware Contractive Embedding (NUCE) loss consistently improved the F1-score across five different transformer backbones, demonstrating its efficacy in handling severely imbalanced datasets. Furthermore, the RF-DETR-based localization module achieved state-of-the-art (SOTA) detection performance for the challenging task of identifying thin, low-contrast CS structures.

Clinical Impact

The development of this automated framework for Cytoplasmic Strings analysis has the potential to significantly improve embryo selection in IVF by providing an objective, non-invasive, and standardized assessment of a critical viability biomarker. This could lead to more accurate embryo grading, increased implantation rates, reduced time to pregnancy, and ultimately, higher success rates and reduced emotional and financial burden for patients undergoing infertility treatment, while reducing reliance on subjective manual inspection.

Limitations

Not explicitly mentioned in the abstract.

Future Directions

Not explicitly mentioned in the abstract, beyond stating that the source code for the framework will be made publicly available.

Medical Domains

Reproductive Medicine Embryology Infertility Treatment Assisted Reproductive Technologies

Keywords

Infertility IVF Embryo Selection Cytoplasmic Strings Time-Lapse Imaging Deep Learning Computer Vision Biomarker

Abstract

Infertility is a major global health issue, and while in-vitro fertilization has improved treatment outcomes, embryo selection remains a critical bottleneck. Time-lapse imaging enables continuous, non-invasive monitoring of embryo development, yet most automated assessment methods rely solely on conventional morphokinetic features and overlook emerging biomarkers. Cytoplasmic Strings, thin filamentous structures connecting the inner cell mass and trophectoderm in expanded blastocysts, have been associated with faster blastocyst formation, higher blastocyst grades, and improved viability. However, CS assessment currently depends on manual visual inspection, which is labor-intensive, subjective, and severely affected by detection and subtle visual appearance. In this work, we present, to the best of our knowledge, the first computational framework for CS analysis in human IVF embryos. We first design a human-in-the-loop annotation pipeline to curate a biologically validated CS dataset from TLI videos, comprising 13,568 frames with highly sparse CS-positive instances. Building on this dataset, we propose a two-stage deep learning framework that (i) classifies CS presence at the frame level and (ii) localizes CS regions in positive cases. To address severe imbalance and feature uncertainty, we introduce the Novel Uncertainty-aware Contractive Embedding (NUCE) loss, which couples confidence-aware reweighting with an embedding contraction term to form compact, well-separated class clusters. NUCE consistently improves F1-score across five transformer backbones, while RF-DETR-based localization achieves state-of-the-art (SOTA) detection performance for thin, low-contrast CS structures. The source code will be made publicly available at: https://github.com/HamadYA/CS_Detection.