Artificial intelligence for prediction of chronic thromboembolic pulmonary hypertension using electrocardiography

Background: Screening and accurate diagnosis of chronic thromboembolic pulmonary hypertension(CTEPH) are critical for managing the progression and preventing associated mortality; however, there are no tools for this purpose. We developed and validated an artificial intelligence (AI) algorithm for predicting CTEPH using electrocardiography (ECG).

Methods: ECG signals were obtained from 54 regular and 23 CTEPH patients to test the technique. A dataset was created by converting the ECG results to digital. The 12-channel ECG signal received from 77 individuals is 924x1300 in size. The end-point was the diagnosis of CTEPH. By applying the suggested Nigerian motif pattern method to this data set, we obtained a feature matrix of 924x15010. FSCmRMR algorithm determined the most influential 947 characteristics among 15010 features and obtained a matrix of 924x947. We used a decision tree, SVM(Support Vector Machine), and KNN(K-Nearest Neighbour )algorithms to classify the selected most weighted features.

Results: We achieved 98.05% success with the decision tree algorithm, 99.89% with the SVM algorithm, and 99.67% with the KNN algorithm. AI algorithm focused on each patient's S-wave, P-wave, and T-wave by QRS complex characteristics.

Conclusion: The AI algorithm demonstrated high accuracy for CTEPH prediction using 12-lead and single-lead ECGs