SPECT-MPI iterative denoising during the reconstruction process using a two-phase learned convolutional neural...

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) faces challenges with image denoising.
• Existing denoising methods can reduce image contrast, impacting diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate a deep neural network-based denoising method integrated into the iterative reconstruction process for SPECT MPI.
• To reduce background coefficient of variation (COV_bkg) and improve contrast-to-noise ratio (CNR) in denoised images.

Main Methods:

• A generative adversarial network (GAN) was employed, trained in two phases: initial training on confined image regions, followed by fine-tuning on full-size images.
• The network was trained and validated using SPECT-MPI data from 247 patients with high- and low-noise scans.

Main Results:

• The proposed iterative deep denoising method significantly reduced COV_bkg by up to 10.28% compared to post-reconstruction low-pass filtering and 12.52% compared to post-reconstruction deep denoising.
• Contrast-to-noise ratio (CNR) was enhanced by up to 54.54% and 45.82% over the same comparison methods, respectively.

Conclusions:

• The iterative deep denoising approach demonstrates superior performance over 2D low-pass Gaussian filtering and post-reconstruction deep denoising methods.
• This technique offers a promising solution for improving image quality in SPECT MPI by preserving contrast while effectively reducing noise.