Deep Learning Model for Real-Time Nuchal Translucency Assessment at Prenatal US

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Prenatal Diagnostics

Background:

• Nuchal translucency (NT) measurement is crucial for early screening of fetal aneuploidies.
• Accurate and efficient NT plane identification and measurement in prenatal ultrasound (US) are essential for reliable screening.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI)-based model for real-time NT plane identification and measurement.
• To assess the AI model's accuracy and consistency compared to experienced radiologists.

Main Methods:

• A retrospective multicenter study developed and evaluated the automated identification and measurement of NT (AIM-NT) model.
• The AIM-NT model's performance was compared against radiologists on internal (3959 images) and external (267 videos) datasets.
• Discrepancies in NT plane identification and thickness measurements were analyzed.

Main Results:

• AIM-NT achieved an area under the receiver operating characteristic curve of 0.92 for NT plane identification on internal data.
• No significant differences were found between AIM-NT and radiologists in NT plane identification accuracy (88.8% vs 87.6%) or NT thickness measurements on external data.
• High consistency was observed in NT plane identification time and thickness measurements (mean difference 0.03 mm).

Conclusions:

• The AIM-NT model demonstrates high accuracy in identifying the NT plane and measuring NT thickness on prenatal US.
• The AI model shows minimal discrepancies with radiologist workflow, suggesting potential for integration into clinical practice.