Enhancing Coronary Artery Disease Prognosis with Artificial Intelligence: A Study on Radionuclide Myocardial Perfusion Imaging

Abstract

Background: Radionuclide myocardial perfusion imaging is widely used as it is a non-invasive method. However, it still requires correlation with invasive procedures such as cardiac catheterization. Nevertheless, there are several different software packages and instruments used. The study of artificial intelligence to assist the nuclear medicine physician for coronary artery disease prognosis will make it more accurate, precise, as well as reduce unnecessary invasive procedures. Objectives: 1. To find the correlation between the results from the nuclear medicine physician and from artificial intelligence 2. To compare the accuracy between the result from artificial intelligence created by different instruments and software packages 3. To study the quantitative value of the different software packages from the one used for generating the artificial intelligence Methods: A retrospective collection of 501 patients was conducted to assess the correlation between nuclear medicine physician reports and artificial intelligence. Quantitative analyses were performed using Corridor 4D and QGSQPS and compared to 4DMSPECT. Additionally, the accuracy of artificial intelligence in SPECT/CT model 870 DR and data reconstructed with QGSQPS and 4DMSPECT operated with SPECT model Infinia were compared. Result: The correlation between the report from the nuclear medicine physicians and artificial intelligence was low (r = -0.006). The correlation between 4DMSPECT and Corridor 4D was between 0.39-0.96, while the Corridor 4D and QGSQPS were between 0.26-0.89. The sensitivity and specificity of the artificial intelligence operated with SPECT model Infinia and data reconstructed with 4DMSPECT were 49.0% and 63.1%, while the one operated with SPECT/CT model 870 DR and data reconstructed with QGSQPS software were 75.6% and 35.3%, respectively. Conclusion: This version of artificial intelligence cannot be used in clinical practice due to its low accuracy and wide range of correlations between software packages. The next improved version will be studied and used in clinical practice.