Related Experiment Videos
Noise correlation in PET, CT, SPECT and PET/CT data evaluated using autocorrelation function: a phantom study on
Pasha Razifar1, Mattias Sandström, Harald Schnieder
1Uppsala University, Centre for Image Analysis, Lägerhyddsv. 3, SE-752 37 Uppsala, Sweden. pasha.razifar@uppsala.imanet.se
BMC Medical Imaging
|August 27, 2005
Summary
Noise patterns in medical imaging (PET, CT, SPECT) were analyzed using autocorrelation functions. Iterative reconstruction methods in PET/CT yield isotropic noise, simplifying analysis, while FBP methods show anisotropic noise for non-circular objects.
Area of Science:
- Medical Imaging
- Image Analysis
- Nuclear Medicine
Background:
- Positron Emission Tomography (PET), Computed Tomography (CT), PET/CT, and Single Photon Emission Tomography (SPECT) are crucial non-invasive imaging techniques.
- These methods generate cross-section images but are susceptible to noise, which can impair image quality and quantitative precision.
- Understanding noise properties, specifically correlation and anisotropy, is vital for accurate image interpretation.
Purpose of the Study:
- To characterize noise correlation and anisotropy in PET, CT, PET/CT, and SPECT images.
- To investigate the impact of reconstruction methods (FBP, OSEM) and imaging parameters on noise properties.
- To evaluate the influence of X-ray dose in PET/CT transmission scans on noise characteristics.
Main Methods:
- Experiments utilized phantoms with varying shapes.
- Data were acquired in 2D mode and reconstructed using Filter Back Projection (FBP) and Ordered Subsets Expectation Maximisation (OSEM) methods.
- Autocorrelation functions (ACF) were employed to analyze noise correlation and variance in different directions.
Main Results:
- Iterative reconstruction in PET/CT produced rotationally symmetric (isotropic) noise, independent of object shape.
- Filter Back Projection (FBP) reconstruction resulted in non-isotropic noise patterns for non-circular objects in PET/CT and CT.
- SPECT images consistently exhibited isotropic noise correlation, regardless of reconstruction algorithm or object shape.
- Noise in PET/CT was unaffected by CT transmission X-ray dose, indicating dominant emission noise.
Conclusions:
- Noise correlation and variance significantly complicate statistical analysis in medical imaging.
- Iterative reconstruction methods are recommended for applications requiring advanced statistical analysis.
- While FBP allows analytical noise calculation, iterative methods do not, presenting a trade-off for statistical analysis.
- Low X-ray doses in the CT transmission component of PET/CT are feasible without compromising image quality or noise behavior.