Page 91 - The-5th-MCAIT2021-eProceeding

& Acharya, 2017) on ECG signals of 40 normal people and 7 CAD patients. Student’s t-test method and
Kruskal–Wallis statistical test are applied to check the extracted features' discrimination ability. The developed
model achieved 0.99% for accuracy, sensitivity, and specificity, respectively.
Table 1. Related Works in Predicting CVDs using ECGs

Approach Dataset Method Feature Output Performance%
Selection
Kiranyaz et al. (2015) MIT-BIH CNN NA Classifying ECG AC = 0.98, SEN =
0.95, SPC = 0.99
Zubair et al. (2016 ) MIT-BIH CNN NA Classifying ECG Ac = 0.93
Acharya et al. Fantasia, CNN NA Diagnosing CAD AC = 94.95, SEN = 93.72,
(2017) St.-Petersburg SEN = 95.18

Kumar et al. (2017) Fantasia, LS-SVM NA Diagnosing CAD AC = 0.99, SEN =
St.-Petersburg 0.99, SPC = 0.99
Tan et al. (2018) PhysioNet LSTM, CNN NA Diagnosing CAD AC = 99.85
(Andersen et al., MIT-BIH CNN, RNN NA Classifying ECG SEN = 0.98, SPC =0.96
2019)

Sharma and Acharya PhysioNet, GSVM OTFC, Identifying CAD AC = 99.53, SEN = 98.64,
(2019) St.Petersburg BWFB SPC = 99.70
(Butun et al., 2020) Fantasia, CapsNet NA Detecting CVD 2sAC=99.44, 5sAC
St.-Petersburg = 98.62
(Al-Zaiti et al., EMPIRE study LR, GBM, NA Predicting CAD SEN =77.0, SPC =
2020) ANN syndrome 76.0
Abbreviations: CNN, Conventional Neural Networks; LS-SVM, Least-Squares Support-Vector Machine; LSTM, Long Short-Term
Memory; RNN, Recurrent Neural Network; GSVM, Gaussian Support Vector Machine; CapNet, Capsule Neural Network; LR,
Logistic Regression; GBM, Gradient Boosting Machine; ANN, Artificial Neural Network; NA, Not Available; OTFC, Optimally
Time-Frequency Concentrated; BWFB, Biorthogonal Wavelet Filter Bank; CAD, Coronary Artery Disease ;AC, Accuracy; SEN,
Sensitivity; SPC, Specificity.

In (Tan et al., 2018), authors implement a long short-term memory (LSTM) network with (CNN) to
automatically diagnose CAD ECG signals. Using Fantasia dataset for normal data and ST-Petersburg for CAD
patients, the model achieved 99.53%, 98.64%, 99.70% accuracy, sensitivity, and specificity, respectively. A
real-time approach for automatic detection of atrial fibrillation (AF) in long-term electrocardiogram (ECG) is
developed in (Andersen, Peimankar, & Puthusserypady, 2019). A combination of CNN and RNN algorithms
trained using MIT-BIH dataset and achieved 98.0% sensitivity and 96.0% specificity. For automatically
identifying CAD, (Sharma & Acharya, 2019) proposed the use of optimally time-frequency concentrated
(OTFC) even-length biorthogonal wavelet filter bank (BWFB). The model was trained on a 10-fold cross-
validation technique and Gaussian Support Vector Machine (GSVM) algorithm to diagnose CAD. The average
sensitivity and specificity obtained are 0.98% and 0.99%, respectively, with the Matthews correlation
coefficient(MCC) of 98.0%. 1D-CADCapsNet (Butun, Yildirim, Talo, Tan, & Acharya, 2020) provides an
automated detection for CAD from ECG signals using Capsule Network algorithm (CapsNet). The proposed
model was trained on two seconds (95,300) and five second-long (38,120) ECG segments from Fantasia and
ST-Petersburg data sets. The 1D-CADCapsNet model yielded a 5-fold diagnosis accuracy of 99.0% and 98.0%
for two and five-second ECG signal groups, respectively. (Al-Zaiti et al., 2020) developed machine learning-
based methods for predicting underlying acute myocardial ischemia in patients with chest pain. The model
trained and tested multiple classifiers on two independent prospective patient cohorts using 554 temporal-spatial
features of the 12-lead ECG.

E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021) [78]
Artificial Intelligence in the 4th Industrial Revolution