Scholarlink

Background <p>The reliability of PET radiomic features is fundamentally constrained by inherent stochastic noise originating from Poisson-distributed photon-counting statistics and additional fluctuations introduced by electronic signal processing. To assess its impact, we repeated PET scans under identical acquisition and reconstruction conditions to evaluate the consistency of radiomic feature measurements and quantify the noise-induced variability as a repeatability measure.</p> Materials and Methods <p>An American College of Radiology-accredited (ACR) phantom was utilized to evaluate measurement variability in PET radiomic features under two scenarios: uniform (Experiment 1) and non-uniform (Experiment 2) radiotracer distributions, employing <sup>18</sup>F-FDG and <sup>68</sup>Ga-PSMA tracers. A total of 93 radiomic features across six categories were extracted using PyRadiomics. Each experiment was repeated ten times, and measurement variability was quantified using the coefficient of variation (CV). Features were benchmarked and classified into three quality grades of repeatability: Grade A (CV &lt; 10%), Grade B (10% ≤ CV &lt; 20%), and Grade C (CV ≥ 20%).</p> Results <p>Experiment 1 identified 22 Grade A and 14 Grade C features, whereas Experiment 2 yielded 30 Grade A and 15 Grade C features. Notably, CV values exhibited substantial variability. In Experiment 1, CV ranged from 0.17% to 205% for <sup>18</sup>F-FDG and from 0.14% to 140% for <sup>68</sup>Ga-PSMA. Similarly, in Experiment 2, CV spanned from 0.16% to 230% for <sup>18</sup>F-FDG and from 0.23% to 79% for <sup>68</sup>Ga-PSMA.</p> Conclusion <p>By isolating inherent stochastic noise, this study applied the coefficient of variation to evaluate measurement variability across repeated PET scans and benchmarked radiomic features into three repeatability quality categories. While some features demonstrated high noise resilience, others showed poor repeatability across repeated scans. These findings highlight the need to establish repeatability standards to guide the development of noise-robust features that support trustworthy imaging biomarkers for clinical applications.</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

Toward Trustworthy PET Radiomics: Quantitative Repeatability Analysis of 18F-FDG and 68Ga-PSMA Features under Inherent Stochastic Noise

  • 期刊论文

  • 发表日期：

    2026年4月8日

查看全文
  • Shu-Ju Tu,
  • Yi-Chen Wu,
  • Jing-Ren Tseng

摘要

Background

The reliability of PET radiomic features is fundamentally constrained by inherent stochastic noise originating from Poisson-distributed photon-counting statistics and additional fluctuations introduced by electronic signal processing. To assess its impact, we repeated PET scans under identical acquisition and reconstruction conditions to evaluate the consistency of radiomic feature measurements and quantify the noise-induced variability as a repeatability measure.

Materials and Methods

An American College of Radiology-accredited (ACR) phantom was utilized to evaluate measurement variability in PET radiomic features under two scenarios: uniform (Experiment 1) and non-uniform (Experiment 2) radiotracer distributions, employing 18F-FDG and 68Ga-PSMA tracers. A total of 93 radiomic features across six categories were extracted using PyRadiomics. Each experiment was repeated ten times, and measurement variability was quantified using the coefficient of variation (CV). Features were benchmarked and classified into three quality grades of repeatability: Grade A (CV < 10%), Grade B (10% ≤ CV < 20%), and Grade C (CV ≥ 20%).

Results

Experiment 1 identified 22 Grade A and 14 Grade C features, whereas Experiment 2 yielded 30 Grade A and 15 Grade C features. Notably, CV values exhibited substantial variability. In Experiment 1, CV ranged from 0.17% to 205% for 18F-FDG and from 0.14% to 140% for 68Ga-PSMA. Similarly, in Experiment 2, CV spanned from 0.16% to 230% for 18F-FDG and from 0.23% to 79% for 68Ga-PSMA.

Conclusion

By isolating inherent stochastic noise, this study applied the coefficient of variation to evaluate measurement variability across repeated PET scans and benchmarked radiomic features into three repeatability quality categories. While some features demonstrated high noise resilience, others showed poor repeatability across repeated scans. These findings highlight the need to establish repeatability standards to guide the development of noise-robust features that support trustworthy imaging biomarkers for clinical applications.