Deep Learning-Based Attenuation Correction Improves Diagnostic Accuracy of Cardiac SPECT

Area of Science:

• Nuclear Medicine
• Cardiology
• Artificial Intelligence in Medical Imaging

Background:

• CT-based attenuation correction (AC) improves myocardial perfusion imaging (MPI) SPECT diagnostic accuracy.
• Existing CT-AC methods are not widely available, increase radiation, and are prone to misregistration.
• There is a need for CT-free AC methods in MPI SPECT.

Purpose of the Study:

• To develop and validate a deep learning model (DeepAC) for generating simulated AC SPECT images from non-AC (NC) SPECT.
• To evaluate the diagnostic accuracy of DeepAC for obstructive coronary artery disease (CAD).
• To compare DeepAC performance against NC and CT-based AC.

Main Methods:

• A conditional generative adversarial neural network (DeepAC) was developed to simulate AC images from NC SPECT data.
• The model was trained on 4,886 NC/AC SPECT studies and validated on 604 external patient studies.
• Diagnostic accuracy for obstructive CAD was assessed using the area under the ROC curve (AUC) for total perfusion deficit (TPD).

Main Results:

• DeepAC images were generated in under 1 second.
• DeepAC TPD showed higher AUC for obstructive CAD (0.79) compared to NC TPD (0.70, P < 0.001), and was similar to AC TPD (0.81).
• DeepAC improved normalcy rates in low-CAD-likelihood patients (70.4%) compared to NC TPD (54.6%, P < 0.001).

Conclusions:

• DeepAC significantly improves diagnostic accuracy for obstructive CAD compared to NC SPECT, achieving accuracy comparable to CT-AC.
• DeepAC obviates the need for CT, reducing radiation exposure and eliminating misregistration artifacts.
• DeepAC offers a rapid, CT-free solution for enhanced MPI SPECT analysis.