Towards Reliable and Reproducible Fetal Brain Biometry: A Deep Learning Approach Using MRI

Fetal brain biometry is a critical component of prenatal care, used to monitor brain development and detect abnormalities. Traditionally, these measurements are performed manually by clinicians, a process that is time-consuming and susceptible to significant inter-observer variability. This study addresses these limitations by developing a fully automated, four-step deep learning framework designed to extract biometric parameters from 3D super-resolution reconstructed fetal brain MRI volumes.

The proposed pipeline integrates a 3D convolutional neural network (CNN) to regress coarse anatomical landmark coordinates from brain segmentation label maps. This is followed by a measurement-specific geometric optimization step that refines these landmark positions by constraining them to relevant anatomical structures. The framework was evaluated using 150 volumes from two publicly available datasets (Zurich and dHCP), covering a gestational age range of 20–37 weeks.

The framework demonstrates strong agreement with ground-truth measurements, achieving mean absolute errors of less than 2 mm for most biometric parameters. Landmark localization accuracy was also high, with mean errors below 4 mm. The study found that the model performed consistently across different datasets and acquisition protocols. When compared to existing automated pipelines, the proposed method showed comparable or improved accuracy. While the geometric refinement step significantly improves the reliability of most measurements, the length of the corpus callosum remains the most difficult parameter to estimate due to its complex anatomical boundaries.

By automating the extraction of biometric measurements, this pipeline reduces the labor-intensive nature of clinical fetal brain assessment and minimizes the risk of human error. The inclusion of an anatomically interpretable landmark-based approach provides clinicians with visual verification of the measurements, which is essential for clinical trust and integration into existing workflows. The release of the implementation to the scientific community supports broader adoption and further research into standardized, reproducible fetal brain biometry.