Scholarlink

Purpose <p>This study evaluates the difference between automatic standardization for the ECDaim platform and expert manual standardization and validates the applicability of the automated method for SPECT brain perfusion images for neurodegenerative diseases.</p> Methods <p>The ECDaim platform is employed to compare automatic and expert manual standardization methods for Tc-99&#xa0;m-ECD SPECT images using data from a local dementia database (390 subjects: NC, AD, LBD, and VaD). All images are subject to spatial normalization and quantitative evaluation using the Hausdorff Distance (HD) and the Dice Coefficient (DC). Whole-brain and small-brain regions, such as the hippocampus, posterior cingulate cortex and precuneus, are analyzed. Differences in uptake ratios for the four brain lobes are measured to determine the performance in terms of quantitative imaging. Statistical parametric mapping (SPM) is used to compare group differences and FWE correction is applied to visualize significant voxel-wise differences between AD and control groups.</p> Results <p>Whole-brain analysis shows that the HD values for automatic and manual standardization methods are less than 5&#xa0;mm for all four groups (NC, AD, LBD, and VaD) and the volume overlap (DC) exceeds 0.95, so there are minimal differences between methods. For small brain regions, HD values are less than 2&#xa0;mm and DC values range from 0.75 to 0.92, with lower values for the PCC. Quantitative measurements of regional uptake ratios for different brain lobes show that the percentage error for all groups is less than 2%. Two-sample t-tests to compare the AD and NC groups show similar spatial distributions for the two standardization methods in terms of voxel-wise statistical results.</p> Conclusion <p>This study uses HD and DC to determine the consistency between automatic and expert manual standardization of Tc-99&#xa0;m-ECD SPECT images and shows that results are similar for both. For the 390 images that are studied, less than 5% are excluded as outliers, so the automatic method is robust and applicable for different disease groups. These results demonstrate that the ECDaim platform unifies the process of imaging analysis and may reduce variations for different manual procedures, improve diagnostic consistency and allow the integration of AI models for clinical decision support.</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

Validation of Automated Standardization Performance for ECDaim Software Developed Using a Taiwan-Specific Database

  • 期刊论文

  • 发表日期：

    2025年3月24日

查看全文
  • Yu-Ching Ni,
  • Hui-Chieh Yang,
  • Ing-Tsung Hsiao,
  • Fan-Pin Tseng,
  • Chia-Yu Lin,
  • Wen-Bin Lin,
  • Zong-Jhe Lyu

摘要

Purpose

This study evaluates the difference between automatic standardization for the ECDaim platform and expert manual standardization and validates the applicability of the automated method for SPECT brain perfusion images for neurodegenerative diseases.

Methods

The ECDaim platform is employed to compare automatic and expert manual standardization methods for Tc-99 m-ECD SPECT images using data from a local dementia database (390 subjects: NC, AD, LBD, and VaD). All images are subject to spatial normalization and quantitative evaluation using the Hausdorff Distance (HD) and the Dice Coefficient (DC). Whole-brain and small-brain regions, such as the hippocampus, posterior cingulate cortex and precuneus, are analyzed. Differences in uptake ratios for the four brain lobes are measured to determine the performance in terms of quantitative imaging. Statistical parametric mapping (SPM) is used to compare group differences and FWE correction is applied to visualize significant voxel-wise differences between AD and control groups.

Results

Whole-brain analysis shows that the HD values for automatic and manual standardization methods are less than 5 mm for all four groups (NC, AD, LBD, and VaD) and the volume overlap (DC) exceeds 0.95, so there are minimal differences between methods. For small brain regions, HD values are less than 2 mm and DC values range from 0.75 to 0.92, with lower values for the PCC. Quantitative measurements of regional uptake ratios for different brain lobes show that the percentage error for all groups is less than 2%. Two-sample t-tests to compare the AD and NC groups show similar spatial distributions for the two standardization methods in terms of voxel-wise statistical results.

Conclusion

This study uses HD and DC to determine the consistency between automatic and expert manual standardization of Tc-99 m-ECD SPECT images and shows that results are similar for both. For the 390 images that are studied, less than 5% are excluded as outliers, so the automatic method is robust and applicable for different disease groups. These results demonstrate that the ECDaim platform unifies the process of imaging analysis and may reduce variations for different manual procedures, improve diagnostic consistency and allow the integration of AI models for clinical decision support.