CTLESS: A scatter-window projection and deep learning-based transmission-less attenuation compensation method for...

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Artificial Intelligence

Background:

• Attenuation compensation (AC) improves SPECT myocardial perfusion imaging (MPI) but often requires CT, increasing radiation and cost.
• SPECT/CT misalignment can lead to diagnostic inaccuracies.

Purpose of the Study:

• To develop and evaluate a CT-less attenuation compensation (CTLESS) method for cardiac SPECT MPI using deep learning.
• To assess the performance of CTLESS compared to CT-based AC (CTAC) and non-AC (NAC) methods.

Main Methods:

• A deep learning network (multi-channel input multi-decoder) was trained to estimate attenuation maps from scatter-energy window photons.
• The estimated maps were segmented, and pre-defined coefficients were assigned for AC.
• Performance was evaluated retrospectively using clinical SPECT/CT MPI data and an anthropomorphic model observer for defect detection.

Main Results:

• CTLESS demonstrated statistically non-inferior performance to CTAC and significantly outperformed NAC for defect detection.
• The method showed comparable performance to CTAC and superior performance to NAC in image quality metrics (RMSE, SSIM).
• CTLESS maintained stable performance with reduced training data and outperformed another deep learning AC method.

Conclusions:

• The CTLESS method provides effective transmission-less AC for SPECT MPI.
• This approach offers a viable alternative to CT-based AC, potentially improving diagnostic accuracy while reducing radiation exposure and costs.