BIBSNet: A deep learning baby image brain segmentation network for MRI scans

BIBSNet:一种用于MRI扫描的深度学习婴儿图像脑部分割网络

阅读:1

Abstract

OBJECTIVES: Brain tissue segmentation of infant magnetic resonance (MR) images is important for studying typical and atypical brain development. The infant brain undergoes rapid changes throughout the first years of postnatal life, making tissue segmentation difficult for most existing algorithms. We introduce a deep neural network BIBSNet (Baby and Infant Brain Segmentation Neural Network), an open-source model for robust and generalizable brain tissue segmentation leveraging data augmentation and a large sample size of manually annotated images. EXPERIMENTAL DESIGN: Model training included MR brain images from 90 participants with an age range of 0-8 months (median age 4.6 months). Using manually annotated real images along with synthetic segmentation images produced using SynthSeg, the model was trained using a 10-fold procedure. Model performance was assessed by comparing BIBSNet, and joint label fusion (JLF) inferred segmentations to ground truth segmentations, and an ad-hoc analysis with iBeat inferred segmentation, using Dice Similarity Coefficient (DSC). Additionally, MR data along with the FreeSurfer compatible segmentations were processed with the DCAN labs infant-ABCD-BIDS processing pipeline from ground truth, JLF, and BIBSNet to produce anatomical and resting state functional derivatives to further assess model performance on processed derivatives. PRINCIPAL OBSERVATIONS: BIBSNet outperforms JLF based on DSC comparisons especially with gray matter (BIBSNet = 0.849, JLF = 0.713) and white matter (BIBSNet = 0.862, JLF = 0.791). Additionally, with processed derived metrics, BIBSNet inferred segmentations outperforms JLF inferred segmentations across nearly all anatomical and functional metrics. Ad-hoc analyses of cortical segmentations - iBeat does not perform subcortical segmentations - showed that there is no significant difference between iBeat and BIBSNet segmentation for infants 0-5 months, but iBeat performed significantly better for infants 6-8 months. CONCLUSIONS: BIBSNet shows marked improvement over JLF across all age groups analyzed. The BIBSNet model is 600x faster compared to JLF at segmentation inference, produces FreeSurfer-compatible segmentation labels, and can be easily included in other processing pipelines. BIBSNet provides a viable alternative for segmenting the brain in the earliest stages of development.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。