Scholarlink

<b>Background:</b> <p>White matter (WM) tract segmentation on diffusion magnetic resonance imaging (dMRI) plays an important role in neurosurgical planning. In spite of many improvements in U-Net architecture, the existing methods fail to perform the delineation of white matter tracts efficiently due to dependence on intermediate computations resulting in errors.</p> <b>Purpose:</b> <p>The multimodal data is used to segment major white matter tracts and study the effect of glioma on white matter using diffusion coefficients.</p> <b>Methods:</b> <p>The attention gating mechanism is employed along with the 3D-Vnet emphasizing on the focussed white matter regions. Each Attention gate(AG) module associated with the decoder part of the Vnet extracts complementary information from the gating signal. The method is applied to pairs of T1-weighted (T1w) and principal direction of diffusion (PDD) maps. The validation is provided on two different dataset viz. the Sheba75 and Human Connectome Project (HCP). The proposed method encompasses the focal loss to handle class imbalance issues and for the efficient prediction of true positives (TP).</p> <b>Results:</b> <p>The proposed AGVnet achieved a mean Dice score of 0.854 on the HCP dataset, demonstrating superior performance over existing segmentation methods such as 3D U-Net and TractSeg. Also, the tract analysis is carried out to study the impact of Glioma tumor through diffusion coefficients such as FA, MD, RD, and AD values in both normal and tumorous patients. The decrease in FA and RD values is observed mainly due to the effect of the tumor on white matter tracts.</p> <b>Conclusion:</b> <p>A novel deep learning-based approach, AGVnet, for accurately delineating white matter tracts on diffusion tensor imaging. The selection of essential features at both the channel and spatial levels leads to the achievement of better segmentation accuracy, especially in complex cases where small and intricate white matter tracts are involved.</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

Unveiling insights: a study on white matter tract segmentation via attention-gate networks and tumor diffusion analysis

  • 期刊论文

  • 发表日期：

    2026年8月1日

查看全文
  • Syed Sajid Hussain,
  • Jainy Sachdeva,
  • Chirag Kamal Ahuja

摘要

Background:

White matter (WM) tract segmentation on diffusion magnetic resonance imaging (dMRI) plays an important role in neurosurgical planning. In spite of many improvements in U-Net architecture, the existing methods fail to perform the delineation of white matter tracts efficiently due to dependence on intermediate computations resulting in errors.

Purpose:

The multimodal data is used to segment major white matter tracts and study the effect of glioma on white matter using diffusion coefficients.

Methods:

The attention gating mechanism is employed along with the 3D-Vnet emphasizing on the focussed white matter regions. Each Attention gate(AG) module associated with the decoder part of the Vnet extracts complementary information from the gating signal. The method is applied to pairs of T1-weighted (T1w) and principal direction of diffusion (PDD) maps. The validation is provided on two different dataset viz. the Sheba75 and Human Connectome Project (HCP). The proposed method encompasses the focal loss to handle class imbalance issues and for the efficient prediction of true positives (TP).

Results:

The proposed AGVnet achieved a mean Dice score of 0.854 on the HCP dataset, demonstrating superior performance over existing segmentation methods such as 3D U-Net and TractSeg. Also, the tract analysis is carried out to study the impact of Glioma tumor through diffusion coefficients such as FA, MD, RD, and AD values in both normal and tumorous patients. The decrease in FA and RD values is observed mainly due to the effect of the tumor on white matter tracts.

Conclusion:

A novel deep learning-based approach, AGVnet, for accurately delineating white matter tracts on diffusion tensor imaging. The selection of essential features at both the channel and spatial levels leads to the achievement of better segmentation accuracy, especially in complex cases where small and intricate white matter tracts are involved.