Prediction of therapy for ischemic heart disease from PPG signals using fuzzy GRU network

Abstract

BackgroundCardiac morbidity like ischemic heart disease (IHD) causes global mortality. Diagnosis of IHD requires coronary angiograms — an invasive, sophisticated, and expensive procedure. However, non-invasive methods could not gain much confidence in swift diagnosis of IHD patients. These potential issues provided the research motivation for diagnosing IHD priorly in this pilot study. ObjectiveA computer-aided technique, as an alternate method to fasten and ease IHD detection and categorization, has been suggested in this cohort study. Here, the classification was conducted by a sandwich model of GRU and fuzzy logic in a deep GRU Fuzzy network. MethodsIn this work, photoplethysmography (PPG) signals were acquired from 355 IHD patients. Gabor–Winger transform was used to derive signal features using statistical identities. Analysis was done with a Fuzzy GRU Network comprising GRU and BiGRU layers with fuzzy layers. An algorithm was designed with a Rete network and three membership functions to deduce fuzzy inference. Then, Choquet Integral was used for defuzzification. ResultsThe proposed network predicted results with 0.85 accuracy, 0.86 recall, 0.84 precision, 0.86 sensitivity, 0.78 Micro-averaged F1-score, and 0.84 specificity. The network architecture was validated using ablation and comparative studies. The PRC and ROC curves gave the highest values of 0.91 AP and 0.90 AUC, respectively for this network. A state-of-the-art study was done to show the superiority of the technique. ConclusionThis computer-aided model using PPG might become a tool in predicting the type of therapy for IHD patients which will provide a non-invasive, inexpensive, and fast mode of diagnosis.