Deep learning-based obstructive coronary artery disease prediction from myocardial perfusion SPECT.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiology and Nuclear Medicine

Background:

• Coronary artery disease (CAD) diagnosis traditionally relies on invasive coronary angiography (ICA).
• Myocardial perfusion SPECT (MP-SPECT) is a non-invasive imaging technique for assessing cardiac function.
• There is a need for improved non-invasive methods to predict obstructive CAD.

Purpose of the Study:

• To apply deep learning (DL) techniques to predict CAD diagnosis from MP-SPECT.
• To evaluate the performance of DL models with different data inputs and attenuation correction methods.
• To assess the impact of incorporating clinical factors on diagnostic accuracy.

Main Methods:

• Retrospective analysis of 515 patients undergoing MP-SPECT.
• Development of DL models for DL-based attenuation correction (DLAC) and CAD prediction (per-patient and per-vessel).
• Training DL models with various inputs: no AC (NAC), DLAC, CT-based AC (CTAC), stress-only, stress/rest, and stress+rest data, with and without clinical factors.

Main Results:

• DLAC improved AUC compared to NAC across datasets and analyses.
• The CTAC-based stress+rest input achieved an AUC of 0.84 for per-patient CAD prediction.
• Incorporating clinical factors (gender, age, hypertension) further increased the AUC to 0.92 for per-patient analysis and 0.80 for per-vessel analysis.

Conclusions:

• DL-based attenuation correction, CTAC, combined stress/rest data, and clinical factors significantly enhance MP-SPECT's predictive performance for CAD.
• DL models offer a promising non-invasive approach for CAD diagnosis, potentially reducing the need for invasive procedures.