Discrimination Between Different Degrees of Coronary Artery Disease Using Time-Domain Features of the Finger Photoplethysmogram in Response to Reactive Hyperemia Publisher



Hosseini ZS¹ ; Zahedi E^{1, 2} ; Movahedian Attar H³ ; Fakhrzadeh H^{4, 5} ; Parsafar MH¹ Show Affiliations
Source: Biomedical Signal Processing and Control Published:2015

Abstract

Atherosclerosis is a major cause of coronary artery disease leading to morbidity and mortality worldwide. Currently, coronary angiography is considered to be the most accurate technique to diagnose coronary artery disease (CAD). However, this technique is an invasive and expensive procedure with risks of serious complications. Since the symptoms of CAD are not noticed until advanced stages of the disease, early and effective diagnosis of CAD is considered a pertinent measure. In this paper, a non-invasive optical signal, the finger photoplethysmogram (PPG) obtained before and after reactive hyperemia is investigated to discriminate between subjects with different CAD conditions. To this end, the PPG from both index fingers and standard 3-lead ECG of 48 patients (16 females, age 54.3 ± 9.6 years and 32 males, age 59.9 ± 10.6 years) scheduled for diagnostic angiography were recorded. The coronary condition of each subject was determined by three expert cardiologists (ground truth) based on these coronary angiograms. Of the 48 patients, 18 were diagnosed as having no disease (normal coronary - NC), 3 were diagnosed as having mild stenosis (MLD), 11 had single-vessel disease (SVD), 5 had two-vessel disease (2VD) and the remaining 11 were reported to have three-vessel disease (3VD). A vessel disease was determined when a significant (more than 50%) stenosis of the lumen cross-sectional area was observed. The 48 subjects were first grouped into two classes, namely high-risk: Class 1 = {2VD, 3VD} (N = 16) and low-risk: Class 2 = {NC, Mild, SVD} (N = 32). Using this approach, classification using a k-Nearest Neighbor classifier leads to an accuracy of 81.5%, a sensitivity of 82.0% and a specificity of 80.9%. Then all 48 subjects were regrouped slightly differently by moving the SVD subjects from the second (low-risk) to the first (high-risk) class. Therefore for the second approach high-risk: Class 1 = {SVD, 2VD, 3VD} (N = 27), whereas low-risk: Class 2 = {NC, Mild} (N = 21). This second approach resulted in an accuracy of 78.8%, a sensitivity of 79.3% and a specificity of 78.3%. We submit that this technique can be employed to implement an efficient triage system for scheduling coronary angiography, as it is able to identify non-invasively patients at greater risk of coronary stenosis. © 2015 Elsevier Ltd. All rights reserved.