Investigation of scatter energy window width and count levels for deep learning-based attenuation map estimation in...

Area of Science:

• Nuclear Medicine
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Deep learning (DL) is crucial for accurate attenuation correction (AC) in cardiac perfusion SPECT imaging.
• Previous studies highlight the benefit of incorporating scatter window images into DL inputs for improved performance.
• The impact of varying scatter window sizes and DL performance at reduced SPECT count levels remain underexplored.

Purpose of the Study:

• To investigate the impact of different scatter window inputs on DL-generated attenuation maps.
• To evaluate the performance of DL for AC in cardiac SPECT imaging acquired at reduced count levels (e.g., quarter-count scans).

Main Methods:

• Utilized a large dataset of 1517 subjects for training, validation, and testing.
• Trained DL models using SPECT images from photopeak and various scatter windows.
• Evaluated DL performance on both full-count and quarter-count SPECT acquisitions.

Main Results:

• Increasing scatter window width (4% to 30%) showed slight improvements in DL-estimated attenuation maps.
• DL models applied to quarter-count SPECT scans exhibited a minor performance reduction compared to full-count scans.
• Discrepancies in performance across scatter window configurations and count levels were minimal (NMSE < 2.1% vs. CT maps).

Conclusions:

• DL demonstrates consistent performance for attenuation map generation across various scatter window settings.
• DL provides accurate attenuation correction even for SPECT scans acquired at a quarter-count level.
• DL shows significant potential for robust AC in cardiac perfusion SPECT, even under reduced data acquisition scenarios.