Fetuses Made Simple: Modeling and Tracking of Fetal Shape and Pose.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computer Vision

Background:

• Fetal MRI analysis is crucial for prenatal diagnostics, but current methods using keypoints or segmentations have limitations.
• Keypoints simplify analysis but miss shape details; segmentations capture shape but complicate motion analysis.

Purpose of the Study:

• To develop a novel 3D articulated statistical fetal body model for improved fetal MRI analysis.
• To overcome limitations of existing methods in capturing both fetal shape and motion dynamics.

Main Methods:

• Constructed a 3D articulated statistical fetal body model using the Skinned Multi-Person Linear Model (SMPL).
• Developed an algorithm for iterative estimation of body pose and shape, enhancing robustness to MRI artifacts.
• Trained the model on a large dataset of 19,816 MRI volumes from 53 subjects.

Main Results:

• The model accurately captures fetal body shape and motion across time series.
• Achieved a surface alignment error of 3.2 mm for 3 mm MRI voxel size on unseen data.
• Enabled automated anthropometric measurements, overcoming limitations of traditional methods.

Conclusions:

• This work presents the first 3D articulated statistical fetal body model.
• The model offers enhanced fetal motion and shape analysis for prenatal diagnostics.
• Provides intuitive visualization and automated measurements, advancing fetal monitoring.