Deep learning designed to read single photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) can improve the diagnosis of coronary artery disease, according to a new international study.
SPECT MPI is commonly used to diagnose coronary artery disease—the most common form of heart disease and killer of more than 370,00 people in the U.S. annually. But to analyze MPI data physicians must combine information available from both the semi-upright and supine view.
And to quantify the data researchers commonly employ a simple rule-based method which calculates the combined total perfusion deficit (TPD), wrote co-author Piotr Slomka, Cedars-Sinai Medical Center in Los Angeles.
“In this study, using a large international cohort of high-efficiency MPI with correlating invasive coronary angiography (ICA), we apply DL to improve the interpretation of upright/supine images,” Slomka et al. added.
The team compared the standard TPD analysis of 1,160 patients with known coronary artery disease against the two-position stress MPI data analyzed by four deep learning methods. Each was trained and evaluated during the validation process.
All patients underwent stress MPI with the radiotracer 99mTc sestamibi and had clinical reads and invasive coronary angiography correlations within six months of MPI. SPECT scanners from four different centers were used and all images were quantified at Cedars-Sinai.
Coronary artery disease was defined as at least 70% narrowing of three major coronary arteries and at least 50% for the left main coronary artery.
Results showed deep learning improved MPI interpretations compared to current methods. A total of 718 (62 percent) patients and 1,272 of 3,480 (37 percent) arteries had obstructive disease. When using deep learning, per-patient sensitivity improved to 65.6%, up from 61.8% with TPD. Per-vessel sensitivity jumped from 54.6% with TPD to 59.1% with deep learning. Deep learning also held a slight edge in sensitivity over on-site clinical readings.
“These findings were demonstrated for the first time in a rigorous, repeated external validation,” Slomka said, in a prepared statement. “Artificial intelligence can be efficiently leveraged to enhance the accuracy of existing nuclear medicine techniques.”
The full study was published in the May issue of the Journal of Nuclear Medicine.