General Purpose Deep Learning Attenuation Correction Improves Diagnostic Accuracy of SPECT MPI: A Multicenter Study.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) relies on CT-based attenuation correction (AC) for diagnostic accuracy.
• Deep learning (DL) offers a potential alternative by generating synthetic AC images.

Purpose of the Study:

• To evaluate if DL-generated synthetic SPECT images can improve the accuracy of conventional SPECT MPI.
• To assess the diagnostic performance of DL-based AC (DeepAC) for obstructive coronary artery disease (CAD).

Main Methods:

• A DL model (DeepAC) was developed and trained on a multicenter cohort of 4,894 patients.
• External validation was performed in two cohorts totaling 1,066 patients from clinical trials.
• Diagnostic accuracy for obstructive CAD was assessed using total perfusion deficit (TPD) and compared with non-attenuation corrected (NC) and CT-based AC (AC).

Main Results:

• DeepAC TPD demonstrated a higher area under the receiver-operating characteristic curve (AUC) (0.77) compared to NC TPD (0.73) in the first external cohort (P < 0.001).
• DeepAC quantitative scores showed better agreement with actual AC scores than NC in the second external cohort.
• The model was validated across multiple sites and patient groups.

Conclusions:

• DeepAC significantly improved prediction performance for obstructive CAD in a multicenter external cohort.
• This DL-based approach can enhance diagnostic accuracy in facilities with conventional SPECT systems.
• DeepAC offers a valuable alternative to CT-based AC without increasing equipment, imaging time, or radiation exposure.