Wavelet technology reduces photon noise in PET imaging

Article

A new wavelet technique developed by researchers in Taiwan promises to rectify the principal downside of PET’s otherwise precise physiological uptake -- low signal to photon ratio due to photon noise.

A new wavelet technique developed by researchers in Taiwan promises to rectify the principal downside of PET's otherwise precise physiological uptake - low signal to photon ratio due to photon noise.

"By processing the image through selected wavelet parameters, we keep the resolution and contrast but reduce almost half of the coefficient of variation in the region of interest of PET images," said Jyh-Cheng Chen, Ph.D., professor of radiological sciences at National Yang-Ming University in Taipei.

A noisy PET image indicates that noise has been overlapped with main signals, according to Chen. Denoising can be achieved through the transfer of signals from spatial domain to wavelet domain using wavelet transform.

Fourier transform can transfer a signal from temporal (spatial) domain to frequency (spatial frequency) domain, Chen said. But because the signal is in sinusoidal form, temporal information is lost after transformation (Comput Med Imaging Graph 2005;29(4):297-304).

"It makes us unable to catch a specific frequency at a certain time interval," Chen said.

However, the basis of the wavelet technique is a waveform of limited duration, so the temporal (spatial) information is retained after transformation. Wavelet transform is able to provide different analytical results by using the benefits of scaling and shifting of the mother wavelet and time-frequency localization, according to Chen.

"This overcomes the shortcoming of Fourier transform and has been used widely to process signals," Chen said.

The denoising procedure is composed of three steps:

  • The wavelet is decomposed into several filters.

  • The row and column of the original image are convoluted via groups of filters. After the decomposition of an image, a threshold is obtained using statistic calculation. By suitable thresholding of wavelet coefficients, the noise can be removed from the image.

  • The row and column of the original image are convoluted via groups of filters. After the decomposition of an image, a threshold is obtained using statistic calculation. By suitable thresholding of wavelet coefficients, the noise can be removed from the image.

"It is quite valuable to spend some effort on threshold selection because it has directly influenced the result of denoising," Chen said. "We processed the data after PET image reconstruction, but the result might have been different - or better - had we done the processing in the sinogram domain."

For more online information, refer to Diagnostic Imaging's PACSweb section.

Newsletter

Stay at the forefront of radiology with the Diagnostic Imaging newsletter, delivering the latest news, clinical insights, and imaging advancements for today’s radiologists.

Recent Videos
SNMMI: Emerging PET Insights on Neuroinflammation with Progressive Apraxia of Speech (PAOS) and Parkinson-Plus Syndrome
Improving Access to Nuclear Imaging: An Interview with SNMMI President Jean-Luc C. Urbain, MD, PhD
SNMMI: 18F-Piflufolastat PSMA PET/CT Offers High PPV for Local PCa Recurrence Regardless of PSA Level
SNMMI: NIH Researcher Discusses Potential of 18F-Fluciclovine for Multiple Myeloma Detection
SNMMI: What Tau PET Findings May Reveal About Modifiable Factors for Alzheimer’s Disease
Emerging Insights on the Use of FES PET for Women with Lobular Breast Cancer
Can Generative AI Reinvent Radiology Reporting?: An Interview with Samir Abboud, MD
Mammography Study Reveals Over Sixfold Higher Risk of Advanced Cancer Presentation with Symptom-Detected Cancers
Combining Advances in Computed Tomography Angiography with AI to Enhance Preventive Care
Study: MRI-Based AI Enhances Detection of Seminal Vesicle Invasion in Prostate Cancer
Related Content
© 2025 MJH Life Sciences

All rights reserved.