Scholarlink

<p>The automatic segmentation of neonatal or newborn brain images is challenging given the paucity of age-specific neonatal atlases and brain templates, image constraints like artifacts and vast differences in the normal maturation patterns of the neonatal brain. To alleviate these concerns, we propose a novel <i>AutoNeoSeg</i> model for the automatic and atlas-free tissue segmentation of neonatal brain scans. Our method performs a five-class tissue delineation of the neonatal brain into gray matter (GM), total white matter (WM), unmyelinated WM (UWM), myelinated WM (MWM) and cerebrospinal fluid (CSF). We design a new contiguity test and use anatomic knowledge constraints for the extraction of brain and CSF. Subsequently, we develop an innovative distance map-based technique for UWM and GM detection. Finally, a novel seed propagation approach is introduced for myelin segmentation. The segmentation results are validated both qualitatively and quantitatively with ground truth images. The proposed algorithm achieves a mean Dice score of 96.9%, 93.9%, 91.4%, 92.2%, 86.8% and 91% for whole brain, GM, total WM, UWM, MWM and CSF segmentation, respectively. The obtained results are superior to other recently published methods in the literature. The major advantages of our <i>AutoNeoSeg</i> model are as follows: it is completely atlas-independent, does not require prior training, performs accurate tissue segmentation especially of the myelinated regions, segments both normal and abnormal scans including brain anomalies and structural deviations, and can be applied to new datasets from a different source without any changes to the existing algorithm. Overall, the proposed work serves as a useful clinical diagnostic tool to study and analyze the developing neonatal brain.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AutoNeoSeg: automatic atlas-free segmentation of neonatal brain MRI

  • 期刊论文

  • 发表日期：

    2025年11月4日

查看全文
  • Chelli N Devi,
  • Anupama Chandrasekharan,
  • V K Sundararaman,
  • Zachariah C Alex

摘要

The automatic segmentation of neonatal or newborn brain images is challenging given the paucity of age-specific neonatal atlases and brain templates, image constraints like artifacts and vast differences in the normal maturation patterns of the neonatal brain. To alleviate these concerns, we propose a novel AutoNeoSeg model for the automatic and atlas-free tissue segmentation of neonatal brain scans. Our method performs a five-class tissue delineation of the neonatal brain into gray matter (GM), total white matter (WM), unmyelinated WM (UWM), myelinated WM (MWM) and cerebrospinal fluid (CSF). We design a new contiguity test and use anatomic knowledge constraints for the extraction of brain and CSF. Subsequently, we develop an innovative distance map-based technique for UWM and GM detection. Finally, a novel seed propagation approach is introduced for myelin segmentation. The segmentation results are validated both qualitatively and quantitatively with ground truth images. The proposed algorithm achieves a mean Dice score of 96.9%, 93.9%, 91.4%, 92.2%, 86.8% and 91% for whole brain, GM, total WM, UWM, MWM and CSF segmentation, respectively. The obtained results are superior to other recently published methods in the literature. The major advantages of our AutoNeoSeg model are as follows: it is completely atlas-independent, does not require prior training, performs accurate tissue segmentation especially of the myelinated regions, segments both normal and abnormal scans including brain anomalies and structural deviations, and can be applied to new datasets from a different source without any changes to the existing algorithm. Overall, the proposed work serves as a useful clinical diagnostic tool to study and analyze the developing neonatal brain.