Scholarlink

<p>Three-dimensional (3D) reconstruction from two-dimensional (2D) medical images is affected by overfitting and blurring of texture, which can compromise the structural accuracy and practicality of reconstruction. This paper presents a texture-aware reconstruction framework that aims to better preserve details and ensure generalization stability. We introduce a Channel Texture Feature with Attention-enabled Decoder (CTF-AtDE) model that combines multi-level feature learning using light coefficient mapping, UNet, and ResNet101, and a specific texture enhancement module. An interactive learning approach based on fission and spatial-channel attention mechanisms is also used to further improve the reconstruction quality and remove blurring artifacts. The experiments on the CQ500 and LIDC-IDRI datasets show that the performance has been improved, and the cross-validation averages of SSIM and PSNR are 0.890 and 0.857, and 32.34 dB and 30.72 dB, respectively. The volumetric consistency of 3D metrics is also very good. This paper shows that attention-guided texture modeling is an effective approach for stable 3D medical image reconstruction.</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

Texture-aware 3D reconstruction via multi-level attention mechanisms and fission-based interactive learning

  • 期刊论文

  • 发表日期：

    2026年5月21日

查看全文
  • Ashish Kumar Gupta,
  • Piyush Kumar

摘要

Three-dimensional (3D) reconstruction from two-dimensional (2D) medical images is affected by overfitting and blurring of texture, which can compromise the structural accuracy and practicality of reconstruction. This paper presents a texture-aware reconstruction framework that aims to better preserve details and ensure generalization stability. We introduce a Channel Texture Feature with Attention-enabled Decoder (CTF-AtDE) model that combines multi-level feature learning using light coefficient mapping, UNet, and ResNet101, and a specific texture enhancement module. An interactive learning approach based on fission and spatial-channel attention mechanisms is also used to further improve the reconstruction quality and remove blurring artifacts. The experiments on the CQ500 and LIDC-IDRI datasets show that the performance has been improved, and the cross-validation averages of SSIM and PSNR are 0.890 and 0.857, and 32.34 dB and 30.72 dB, respectively. The volumetric consistency of 3D metrics is also very good. This paper shows that attention-guided texture modeling is an effective approach for stable 3D medical image reconstruction.