Integrated artificial intelligence-based classification approach for prediction of acute coronary syndrome

Coronary heart disease (CHD) is one of the most life-threatening diseases all over the world. One of the common medical emergencies and the leading cause of hospitalization, morbidity and mortality is known as acute coronary syndrome (ACS). ACS, which refers to a wide range of acute myocardial ischaemic conditions, is a dynamic and unstable process. The failure in timely diagnosis and prompt treatment of ACS may lead to fatal outcomes. The existing gap of knowledge in “timely and accurate” diagnosis and classification of the patients with suspected ACS is an extremely challenging issue for the practicing emergency physicians. Therefore, application of an innovative approach is required. Using “AI-based” classification models could be considered as an innovative, creative, and multi-disciplinary strategy. Recently, the “hybrid AI-based” classification models have gained more attraction due to the inefficiency of conventional “single AI-based” models in accurate classification. Accordingly, the present study attempts to introduce a novel hybrid classification model for the prediction of ACS to fill the multi-stages gaps. To this end, as the initial stage, the pros and cons of the “single AI-based” were evaluated toward providing a strategy in development of the best classification models for prediction of heart failure based on the Perth data set. In the second stage, a registry entitled “Acute Coronary Syndrome Event — in Kermanshah, Iran (ACSEKI)” was designed and established as the first ACS registry in Iran. The following results were obtained when classification of the ACS types used the conventional “single AI-based” methods. The comparison results of the classifiers showed the highest accuracy of 83.2% and 82.9% for the Feed-forward backpropagation neural network (FFBPNN) and K-nearest neighbors (K-NNs) methods respectively. Although FFBPNN classifier is slightly more accurate than K-NN, there are some advantages such as simple implementability, understandability and interpretability for the latter. In the development of the “hybrid AI-based” classification models, the proposed model (K1-K2- NN), was basically introduced through combining AI approaches of modified K-NN, genetic algorithm (GA), Fisher’s discriminant ratio (FDR) and class separability criteria (CSC). The classification performance of K1-K2-NN model was benchmarked against 13 commonly used classification models using repeated random sub-sampling crossvalidation on ACSEKI data set. The optimized K1-K2-NN model (3-5-NN) demonstrated higher performance accuracy with an average of 94.4% ± 0.9%. As the core component of the present study, the previous models were improved by introducing a “Developed Feed Forward Back Propagation Neural Network” (DFFBPNN). Performance evaluation of the proposed model were conducted by comparing 13 well-known classification models based on various commonly used evaluation criteria on seven data sets (ACSEKI data set as well as six data sets taken from the University of California Irvine (UCI) machine-learning repository). Statistical analysis was performed using the Friedman test followed by post-hoc tests. Finally, the performance results of the proposed model was benchmarked against the best ones reported as the state-of-the-art classifiers in terms of classification accuracy for the same data sets. The experimental findings indicated that the novel proposed hybrid model resulted in significantly better classification performance compared with all 13 classification methods. The classification accuracy of the “hybrid model” and “K1-K2-NN” on ACSEKI data set were 95.2% and 94.2%, respectively, showing 0.08% improvement for the “hybrid model”. Furthermore, substantial findings of the comprehensive comparative study revealed that performance of the proposed model in terms of classification accuracy is desirable, promising, and competitive to the existing state-of-the-art classification models. Accordingly, the proposed “hybrid model” demonstrated to be applicable for classification problems in different medical areas, particularly for early detection of ACS. To recapitulate, the study demonstrated that an integrated AI-based classification approach could be a significant potential for prediction of ACS. Thus, the proposed model could be effectively used for a clinician with less experience or as a second opinion for an experienced senior clinician to their quickly, timely, and accurately decision making process. The model could also be utilized for classification tasks in the other medical fields such as breast cancer and diabetes. This model is expected to make a significant contribution to the literature of integrated AI-based approach for classification of ACS with high accuracy and efficiency.

Preview Text fs 2014 73 ir.pdf Download (2MB) | Preview

Actions (login required)

View Item