Scholarlink

<p>Computed Tomography (CT) is a critical imaging modality in modern medical diagnostics, providing multi-planar structural information and high-resolution 3D volumetric data. However, conventional CT typically requires high-dose, multi-angle acquisitions to ensure image quality, which increases the risk of radiation exposure to patients. To address this issue, this study proposes a two-stage 3D reconstruction framework that combines super-resolution enhancement with neural radiance field learning, aiming to reconstruct 3D-aware CT projection sequences from single-view, low-resolution X-ray images after multi-view supervised training, thereby reducing reliance on high-dose and multi-view acquisitions. In the first stage, a lightweight super-resolution network, CA-ELAN, is designed, which enhances spatial features and texture details through coordinate attention mechanisms. In the second stage, the SESA-MedNeRF model is employed, integrating squeeze-and-excitation and spatial attention to learn continuous 3D volumes from the enhanced images, effectively capturing anatomical structures and density variations. Experiments on public knee and chest X-ray datasets demonstrate that the proposed method achieves excellent structural recovery, texture clarity, and volumetric consistency under single-view and low-dose conditions, outperforming several NeRF baseline models in both quantitative and qualitative assessments. This study confirms that the combination of super-resolution guidance and neural radiance field modeling can effectively reduce radiation exposure while maintaining high reconstruction quality, showing significant clinical potential and offering new insights for the advancement of low-dose, high-precision medical imaging technologies.</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

Dual-Phase attention cascade network: From low-resolution X-ray enhancement to high-fidelity 3D-aware CT-projections

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Yan Wang,
  • Yuquan Wang,
  • Xinbo Wang,
  • Yongzheng Tan

摘要

Computed Tomography (CT) is a critical imaging modality in modern medical diagnostics, providing multi-planar structural information and high-resolution 3D volumetric data. However, conventional CT typically requires high-dose, multi-angle acquisitions to ensure image quality, which increases the risk of radiation exposure to patients. To address this issue, this study proposes a two-stage 3D reconstruction framework that combines super-resolution enhancement with neural radiance field learning, aiming to reconstruct 3D-aware CT projection sequences from single-view, low-resolution X-ray images after multi-view supervised training, thereby reducing reliance on high-dose and multi-view acquisitions. In the first stage, a lightweight super-resolution network, CA-ELAN, is designed, which enhances spatial features and texture details through coordinate attention mechanisms. In the second stage, the SESA-MedNeRF model is employed, integrating squeeze-and-excitation and spatial attention to learn continuous 3D volumes from the enhanced images, effectively capturing anatomical structures and density variations. Experiments on public knee and chest X-ray datasets demonstrate that the proposed method achieves excellent structural recovery, texture clarity, and volumetric consistency under single-view and low-dose conditions, outperforming several NeRF baseline models in both quantitative and qualitative assessments. This study confirms that the combination of super-resolution guidance and neural radiance field modeling can effectively reduce radiation exposure while maintaining high reconstruction quality, showing significant clinical potential and offering new insights for the advancement of low-dose, high-precision medical imaging technologies.