Scholarlink

The thalamus is a brain structure of great interest regarding neurodegenerative diseases. The correct estimation of its shape and nuclei volumes can lead to a successful application of some kind of therapies for the patients. However, the correct delineation of thalamic nuclei from resonance imaging (MRI) requires the use of White-Mater Nulled (WMn) images, which are not common. Current segmentation approaches synthetize WMn-like images from T1-weighted MRI, and use these synthetized MRI along with the original ones to segment the thalamic nuclei. However, the creation of WMN-like images from T1 is not always as good as needed, and segmentations can fail. Recently, the HIPS-THOMAS approach has been proposed, incorporating a third order polynomic function as a very fast and simple WMn-like synthesizer (HIPS) along with the atlas-based approach THOMAS. This new method has achieved outstanding segmentations from T1-W MRI, but the inclusion of HIPS with other approaches like convolutional neural networks has not been explored. This work uses convolutional neural networks (CNN) for segmenting thalamic nuclei from T1-w MRI combined with the third-order polynomic function used by HIPS-THOMAS. The tests performed show that the combination of both tools can produce accurate thalamic nuclei estimations, using only T1-w MRI.

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

Automated Thalamic Nuclei Segmentation from Brain T1-W MRI Using Convolutional Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Robin Cabeza-Ruiz,
  • Luis Velázquez-Pérez,
  • Alejandro Linares-Barranco,
  • Roberto Pérez-Rodríguez

摘要

The thalamus is a brain structure of great interest regarding neurodegenerative diseases. The correct estimation of its shape and nuclei volumes can lead to a successful application of some kind of therapies for the patients. However, the correct delineation of thalamic nuclei from resonance imaging (MRI) requires the use of White-Mater Nulled (WMn) images, which are not common. Current segmentation approaches synthetize WMn-like images from T1-weighted MRI, and use these synthetized MRI along with the original ones to segment the thalamic nuclei. However, the creation of WMN-like images from T1 is not always as good as needed, and segmentations can fail. Recently, the HIPS-THOMAS approach has been proposed, incorporating a third order polynomic function as a very fast and simple WMn-like synthesizer (HIPS) along with the atlas-based approach THOMAS. This new method has achieved outstanding segmentations from T1-W MRI, but the inclusion of HIPS with other approaches like convolutional neural networks has not been explored. This work uses convolutional neural networks (CNN) for segmenting thalamic nuclei from T1-w MRI combined with the third-order polynomic function used by HIPS-THOMAS. The tests performed show that the combination of both tools can produce accurate thalamic nuclei estimations, using only T1-w MRI.