Scholarlink

<p>Diffusion tensor imaging (DTI) provides valuable microstructural information through 3D diffusion measurements, with its accuracy largely dependent on the number of diffusion-encoding gradient directions (NDGD). While higher NDGD improves estimates of metrics such as mean diffusivity (MD) and Fractional Anisotropy (FA), it also increases scan time and computational burden. To address this, we propose a U-Net-based model that integrates Inception modules for multiscale feature extraction and Convolutional Block Attention Modules (CBAM) for refining spatial and channel information. The model generates high-quality, increased NDGD-equivalent data from reduced NDGD, thereby preserving critical microstructural details while significantly reducing scan time. The experimental results demonstrate strong performance, with an SSIM of 99.75%, PSNR of 55.9&#xa0;dB, and low Artifact Power (0.14). Furthermore, FA and MD maps derived from the model closely matched those from full NDGD, confirming its robustness. These findings highlight the potential of the proposed approach to make DTI more efficient and clinically practical by shortening scans without compromising diagnostic accuracy or patient comfort.</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

Optimizing Diffusion Tensor Imaging (DTI) via Deep Learning with Minimal Number of Diffusion Encoding Gradient Directions (NDGD)

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Sarah Saeed Khan,
  • Irfan Ullah,
  • Faisal Najeeb,
  • Hammad Omer

摘要

Diffusion tensor imaging (DTI) provides valuable microstructural information through 3D diffusion measurements, with its accuracy largely dependent on the number of diffusion-encoding gradient directions (NDGD). While higher NDGD improves estimates of metrics such as mean diffusivity (MD) and Fractional Anisotropy (FA), it also increases scan time and computational burden. To address this, we propose a U-Net-based model that integrates Inception modules for multiscale feature extraction and Convolutional Block Attention Modules (CBAM) for refining spatial and channel information. The model generates high-quality, increased NDGD-equivalent data from reduced NDGD, thereby preserving critical microstructural details while significantly reducing scan time. The experimental results demonstrate strong performance, with an SSIM of 99.75%, PSNR of 55.9 dB, and low Artifact Power (0.14). Furthermore, FA and MD maps derived from the model closely matched those from full NDGD, confirming its robustness. These findings highlight the potential of the proposed approach to make DTI more efficient and clinically practical by shortening scans without compromising diagnostic accuracy or patient comfort.