<p>Positron emission tomography/computed tomography (PET/CT) is widely used to diagnose lung cancer. Accurate segmentation of lung cancer lesions in PET/CT images is critical for diagnosis and treatment planning. However, Poisson–Gaussian noise generated during image acquisition degrades image quality, particularly at lesion edges, reducing segmentation accuracy. The median-modified Wiener filter (MMWF) effectively reduces noise while preserving structural details; however, its performance highly depends on kernel size. This study aims to optimize the MMWF algorithm kernel size and evaluate its usefulness in reducing noise and improving cancer segmentation performance in lung cancer PET/CT images. We added Poisson–Gaussian noise with standard deviation of 0.001, 0.002, and 0.004 to PET/CT images to simulate low-, moderate-, and high-noise conditions, respectively, and compared the MMWF algorithm against conventional median and Wiener filters as baselines. The MMWF algorithm was optimized by increasing the kernel size from 3 <InlineEquation ID="IEq1"> <EquationSource Format="TEX">$\times$</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> 3 to 13 <InlineEquation ID="IEq2"> <EquationSource Format="TEX">$\times$</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> 13 in 2 <InlineEquation ID="IEq3"> <EquationSource Format="TEX">$\times$</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> 2 increments. Quantitative evaluation results showed that the 9 <InlineEquation ID="IEq4"> <EquationSource Format="TEX">$\times$</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> 9 kernel size achieved the highest segmentation and image quality performance across all evaluation metrics and noise levels. Statistical analysis using linear mixed-effects models confirmed that the optimized MMWF 9 <InlineEquation ID="IEq5"> <EquationSource Format="TEX">$\times$</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> 9 demonstrated significantly superior performance compared with both baseline filters across all metrics (all <i>p</i> < 0.001). In conclusion, the optimized MMWF kernel size effectively reduces noise and improves segmentation accuracy in PET/CT images, highlighting the potential applicability of the MMWF algorithm in PET/CT-based lung cancer analysis.</p>

Optimized median-modified Wiener filter-based segmentation of lung cancer in PET/CT images

期刊论文
发表日期：

2026年7月24日

查看全文

Hajin Kim,
Sewon Lim,
Youngjin Lee

摘要

Positron emission tomography/computed tomography (PET/CT) is widely used to diagnose lung cancer. Accurate segmentation of lung cancer lesions in PET/CT images is critical for diagnosis and treatment planning. However, Poisson–Gaussian noise generated during image acquisition degrades image quality, particularly at lesion edges, reducing segmentation accuracy. The median-modified Wiener filter (MMWF) effectively reduces noise while preserving structural details; however, its performance highly depends on kernel size. This study aims to optimize the MMWF algorithm kernel size and evaluate its usefulness in reducing noise and improving cancer segmentation performance in lung cancer PET/CT images. We added Poisson–Gaussian noise with standard deviation of 0.001, 0.002, and 0.004 to PET/CT images to simulate low-, moderate-, and high-noise conditions, respectively, and compared the MMWF algorithm against conventional median and Wiener filters as baselines. The MMWF algorithm was optimized by increasing the kernel size from 3 $\times$ $\times$ 3 to 13 $\times$ $\times$ 13 in 2 $\times$ $\times$ 2 increments. Quantitative evaluation results showed that the 9 $\times$ $\times$ 9 kernel size achieved the highest segmentation and image quality performance across all evaluation metrics and noise levels. Statistical analysis using linear mixed-effects models confirmed that the optimized MMWF 9 $\times$ $\times$ 9 demonstrated significantly superior performance compared with both baseline filters across all metrics (all p < 0.001). In conclusion, the optimized MMWF kernel size effectively reduces noise and improves segmentation accuracy in PET/CT images, highlighting the potential applicability of the MMWF algorithm in PET/CT-based lung cancer analysis.