Scholarlink

<p>The reconstruction of volumetric CT structures from two-dimensional (2D) slices is still a challenging task due to the lack of spatial information and the loss of structure details. In this paper, we introduce a texture-aware framework for three-dimensional (3D) CT reconstruction using the Channel Texture Enhancement Model (CTEM). In this framework, we design a dual-branch encoder using the combination of UNet and ResNet101 and incorporate the proposed Color-Invariant Light Coefficient Mapping (CI-LCM) and texture enhancement using Channel Binary Pattern (CBP), Channel Directional Pattern (CDP), and Channel Orientation Pattern (COP) descriptors to achieve more consistent reconstruction in terms of structure and texture. Moreover, we use the proposed Fission Based Interactive Learning (FBIL) strategy, which combines the Teaching-Learning-Based Optimization (TLBO), Spider Monkey Optimization (SMO), and Monkey King Optimization (MKO) algorithms to improve the learning performance of the model. The experimental results on the CQ500 dataset and LIDC-IDRI dataset using the proposed approach have shown that the proposed approach outperforms the recent state-of-the-art methods, including DVAOM, LRM, and D3T, in terms of reconstruction quality. In the CQ500 dataset, the proposed model achieves the 3D IoU of 0.868, PSNR of 30.06 dB, LPIPS of 0.1019, and SSIM of 0.887. Ablation studies and five-fold family-wise cross-validation have validated the proposed approach for volumetric CT 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

A texture-rich, color-invariant 3D medical image reconstruction utilizing fission-based interactive learning optimization

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Ashish Kumar Gupta,
  • Piyush Kumar

摘要

The reconstruction of volumetric CT structures from two-dimensional (2D) slices is still a challenging task due to the lack of spatial information and the loss of structure details. In this paper, we introduce a texture-aware framework for three-dimensional (3D) CT reconstruction using the Channel Texture Enhancement Model (CTEM). In this framework, we design a dual-branch encoder using the combination of UNet and ResNet101 and incorporate the proposed Color-Invariant Light Coefficient Mapping (CI-LCM) and texture enhancement using Channel Binary Pattern (CBP), Channel Directional Pattern (CDP), and Channel Orientation Pattern (COP) descriptors to achieve more consistent reconstruction in terms of structure and texture. Moreover, we use the proposed Fission Based Interactive Learning (FBIL) strategy, which combines the Teaching-Learning-Based Optimization (TLBO), Spider Monkey Optimization (SMO), and Monkey King Optimization (MKO) algorithms to improve the learning performance of the model. The experimental results on the CQ500 dataset and LIDC-IDRI dataset using the proposed approach have shown that the proposed approach outperforms the recent state-of-the-art methods, including DVAOM, LRM, and D3T, in terms of reconstruction quality. In the CQ500 dataset, the proposed model achieves the 3D IoU of 0.868, PSNR of 30.06 dB, LPIPS of 0.1019, and SSIM of 0.887. Ablation studies and five-fold family-wise cross-validation have validated the proposed approach for volumetric CT reconstruction.