ABSTRACT
Background and Purpose
Deep learning enables fast fetal brain age prediction from MRI. However, most models emphasize global features while ignoring local edge details. To improve accuracy, we propose a novel model incorporating global edge information, achieving performance comparable to that of experienced clinicians.
Methods
A retrospective collection of 1630 fetal brain coronal T2-weighted MR images from 207 singleton pregnancies with brain ages ranging from 22 to 38 weeks from June 2019 to July 2023 was performed. The fetal MRI dataset was divided into two independent subsets: a training dataset for optimizing model parameters and a test dataset for evaluating the model’s performance relative to a reference standard. Four-fifths of the dataset were allocated as training data and one-fifth as test data. We trained a neural network that incorporates global edge information and continuously optimized it using performance indicators such as mean absolute error (MAE) and r-square (R
2) to achieve the desired results.
Results
In a retrospective study involving 1630 fetal brain MR images from 207 subjects, the edge-guided deep learning model achieved higher accuracy in predicting fetal age compared to existing methods, with an MAE of 0.79 weeks and an R
2 value of 0.94, and can promote regression models to produce more stable and reliable predictions.
Conclusions
Compared to existing methods for predicting fetal brain age, our method demonstrates superior performance and is helpful for accurately assessing the fetal brain development in clinical practice.
If you do not see content above, kindly GO TO SOURCE.
Not all publishers encode content in a way that enables republishing at Neuro.vip.
This post is Copyright: Haitao Gan,
Qingsong Gao,
Zhi Yang,
Yiyuan Zhou,
Ran Zhou,
Yu Guo,
Wei Xia | November 4, 2025
Wiley: Journal of Neuroimaging: Table of Contents