Decoding fetal motion in 4D ultrasound with DeepLabCut

Area of Science:

• Biomedical Engineering
• Developmental Neuroscience
• Medical Imaging

Background:

• Fetal motor behavior is crucial for assessing neurodevelopment.
• Current qualitative assessments lack precision.
• Quantitative analysis of fetal movements is needed.

Purpose of the Study:

• To objectively analyze fetal motor behavior using DeepLabCut (DLC) and four-dimensional ultrasound (4DUS).
• To develop a novel clinical method for precise fetal neurodevelopment assessment.
• To quantitatively evaluate fetal movement patterns and coordination.

Main Methods:

• Fifty 4DUS recordings (12-22 weeks gestation) were analyzed.
• A custom DLC model was trained using manually labeled fetal joints.
• Kinematic analysis quantified movement patterns and developmental coordination.

Main Results:

• DLC demonstrated high tracking accuracy (mean likelihood score 0.960).
• Manual labeling showed excellent reproducibility (inter-rater ICC=0.990).
• Movement patterns evolved from localized limb activity to coordinated systemic movements, indicating neuromuscular maturation.

Conclusions:

• DLC enables precise, quantitative analysis of fetal motor behavior from 4DUS.
• This AI-driven method is a noninvasive alternative to qualitative assessments.
• It offers insights into neurodevelopmental trajectories and potential early screening for disorders.