Related Experiment Video
Updated: Sep 21, 2025
10:25
Deep Learning-Based Segmentation of Cryo-Electron Tomograms
Published on: November 11, 2022
9.4K
ID-Seg: an infant deep learning-based segmentation framework to improve limbic structure estimates
Yun Wang1,2, Fateme Sadat Haghpanah3, Xuzhe Zhang4
1Department of Psychiatry and Behavioral Sciences, Duke University, Durham, NC, USA.
Brain Informatics
|May 28, 2022
Summary
We developed Infant Deep learning SEGmentation Framework (ID-Seg) for accurate infant brain MRI segmentation. ID-Seg improves accuracy over existing methods and shows stronger associations with infant behavioral problems.
Area of Science:
- Neuroimaging
- Developmental Neuroscience
- Artificial Intelligence in Medicine
Background:
- Infant brain magnetic resonance imaging (MRI) is crucial for studying early neurodevelopment.
- Segmenting small limbic structures in infant MRI is challenging due to low contrast and high curvature.
- Current deep learning models for infant MRI segmentation often use small datasets, risking generalization issues.
Purpose of the Study:
- To develop and validate a robust deep learning framework for segmenting limbic structures in infant brain MRI.
- To compare the performance of the new framework against established methods like the Developmental Human Connectome (dHCP) pipeline.
- To assess the clinical relevance of segmentations by linking them to infant behavioral outcomes.
Main Methods:
- Leveraged a large infant MRI dataset (n=473) and transfer learning to pre-train a deep convolutional neural network.
- Developed the Infant Deep learning SEGmentation Framework (ID-Seg) using a leave-one-out cross-validation strategy for fine-tuning.
- Evaluated ID-Seg on two independent datasets and compared its performance with the dHCP pipeline using Dice Similarity Coefficient (DSC), Intra-class Correlation (ICC), and Average Surface Distance (ASD).
Main Results:
- ID-Seg achieved high segmentation accuracy with a mean DSC of 0.87, ICC of 0.93, and ASD of 0.31 mm across datasets.
- ID-Seg significantly improved segmentation accuracy compared to the dHCP pipeline.
- Estimates from ID-Seg showed stronger associations with behavioral problems in infants at age 2 compared to dHCP estimates.
Conclusions:
- ID-Seg offers a robust and accurate method for segmenting amygdala and hippocampus in infant brain MRI.
- The framework demonstrates superior performance and clinical relevance compared to existing pipelines.
- Future work should focus on multi-site validation and extending ID-Seg to other brain regions.
