Deep Learning Based Junctional Zone Quantification using 3D Transvaginal Ultrasound in Assisted Reproduction

Area of Science:

• Reproductive Medicine
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• The Uterine Junctional Zone (JZ) is crucial for embryo implantation in assisted reproduction.
• Current manual assessment of the JZ is time-consuming, operator-dependent, and subjective.
• Objective quantification of JZ is needed to improve assisted reproduction outcomes.

Purpose of the Study:

• To develop and validate automated methods for JZ visualization and quantification using 3D-TVUS.
• To assess the performance of automated JZ analysis compared to manual methods.
• To enhance the efficiency and objectivity of JZ assessment in clinical practice.

Main Methods:

• A deep learning algorithm pipeline was developed to process 3D-TVUS uterus acquisitions.
• The algorithm generates a point cloud of the endometrium surface to render the midcoronal plane.
• The automated system is designed to handle uterine structural variations and pathologies.

Main Results:

• The automated JZ assessment demonstrated reliable performance in over 89% of cases evaluated by an expert clinician.
• The automated solution reproduced and sometimes surpassed manual workflow accuracy.
• Clinical workflow efficiency was improved approximately tenfold with reduced operator bias.

Conclusions:

• Automated 3D-TVUS analysis provides an efficient and objective method for Uterine Junctional Zone assessment.
• This technology has the potential to improve diagnostic accuracy and patient outcomes in assisted reproduction.
• Further integration of AI in medical imaging can overcome limitations of manual analysis.