Post-reconstruction attenuation correction for SPECT myocardium perfusion imaging facilitated by deep learning-based...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Nuclear Medicine

Background:

• Single-photon emission computed tomography (SPECT) images often lack quantitative accuracy due to attenuation artifacts.
• Existing SPECT-only systems typically produce non-attenuation corrected (NC) images, limiting diagnostic precision.
• Developing effective attenuation correction (AC) methods is crucial for improving SPECT image quality.

Purpose of the Study:

• To develop and evaluate a post-reconstruction attenuation correction (PRAC) approach for myocardial perfusion SPECT imaging.
• To utilize a deep learning (DL) based attenuation map to improve quantitative accuracy in SPECT images.
• To enable attenuation correction for SPECT-only systems without requiring CT data.

Main Methods:

• A novel PRAC method was developed using a deep learning convolutional neural network to generate attenuation maps from NC SPECT images.
• Projection data were re-estimated by forwardly projecting the NC image.
• The DL-generated attenuation map was integrated into an offline reconstruction algorithm to produce attenuation-corrected (AC) images.

Main Results:

• The PRAC method with DL-generated attenuation maps showed high consistency with vendor-generated AC images.
• Mean absolute errors were comparable to CT-based methods, with low errors in the left ventricular (LV) blood pool and myocardium.
• Summed stress scores derived from PRAC images were more accurate compared to NC images.

Conclusions:

• A feasible PRAC approach using DL-based attenuation maps was successfully developed for myocardial perfusion SPECT imaging.
• This method holds potential for generating attenuation-corrected images from SPECT-only scanner data.
• The DL-driven PRAC technique offers a promising solution for enhancing quantitative accuracy in SPECT imaging.