Improved diagnostic accuracy for myocardial perfusion imaging using artificial neural networks on different input variables including clinical and quantification data

Abstract Objective Diagnostic accuracy of myocardial perfusion imaging (MPI) is not optimal to predict the result of angiography. The current study aimed at investigating the application of artificial neural network (ANN) to integrate the clinical data with the result and quantification of MPI. Methods Out of 923 patients with MPI, 93 who underwent angiography were recruited. The clinical data including the cardiac risk factors were collected and the results of MPI and coronary angiography were recorded. The quantification of MPI polar plots (i.e. the counts of 20 segments of each stress and rest polar plots) and the Gensini score of angiographies were calculated. Feed-forward ANN was designed integrating clinical and quantification data to predict the result of angiography (normal vs. abnormal), non-obstructive or obstructive coronary artery disease (CAD), and Gensini score (≥10 and Results The accuracy of MPI to predict the result of angiography, obstructive CAD, and Gensini score increased from 81.7% to 92.9%, 65.0% to 85.7%, and 50.5% to 92.9%, respectively by ANN using counts and clinical risk factors. Conclusion The diagnostic accuracy of MPI could be improved by ANN, using clinical and quantification data.