Scholarlink

3D reconstruction from 2D medical images is a challenging approximation method due to the unavailability of point normal. Medical 3D models are generated using Marching Cube (MC) algorithm but it generates a rough and ugly surface mesh. Mesh denoising is done to smooth the noise using many techniques such as Laplacian, Humphrey’s Class (HC), Mean Curvature Flow (MCF), Taubin’s signal processing approach, and Scale Dependant Umbrella (SDU). In this paper, an improved scale dependant umbrella mesh denoising algorithm (ISDUMD) is proposed that corrects the SDU algorithm up to 4.21% using the Bayesian Averaging Method (BAM). BAM is used to determine the average of a distribution from a predetermined average. The constant ‘c’ in the BAM is experimentally determined at 0.001. The proposed ISDUMD (SDU + BAM) method had 4.03\% less volume shrinkage as compared to SDU and 7.12% better Structural Similarity Index Measure (SSIM) score than SDU after 10 iterations. The test is done on a brain model with synthetically added 33.217 dB PSNR random noise. The dataset used is UPENN-GBM and is openly available.

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

An ISDUMD Algorithm Using Bayesian Averaging for Smoothing 3D Reconstruction of 2D MRI Medical Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mriganka Sarmah,
  • Arambam Neelima,
  • Puspakshi Sarmah

摘要

3D reconstruction from 2D medical images is a challenging approximation method due to the unavailability of point normal. Medical 3D models are generated using Marching Cube (MC) algorithm but it generates a rough and ugly surface mesh. Mesh denoising is done to smooth the noise using many techniques such as Laplacian, Humphrey’s Class (HC), Mean Curvature Flow (MCF), Taubin’s signal processing approach, and Scale Dependant Umbrella (SDU). In this paper, an improved scale dependant umbrella mesh denoising algorithm (ISDUMD) is proposed that corrects the SDU algorithm up to 4.21% using the Bayesian Averaging Method (BAM). BAM is used to determine the average of a distribution from a predetermined average. The constant ‘c’ in the BAM is experimentally determined at 0.001. The proposed ISDUMD (SDU + BAM) method had 4.03\% less volume shrinkage as compared to SDU and 7.12% better Structural Similarity Index Measure (SSIM) score than SDU after 10 iterations. The test is done on a brain model with synthetically added 33.217 dB PSNR random noise. The dataset used is UPENN-GBM and is openly available.