การประยุกต์ใช้โครงข่ายประสาทเทียมแบบสังวัตนาการสำหรับการตรวจหาภาวะความเสี่ยงการเป็นโรคหัวใจ

Abstract

Arrhythmias is one of the leading causes of death in the world, occurring from genetic or unhealthy lifestyle activities. According to the study of the structure of the heart system, it was found that there are four parts of the heart which consist of SA Node, AV Node, Bundle Branches, and Purkinje fiber. We could detect heart disease signals by EKG. Moreover, this thesis uses EKG samples from Physionet.org which totally of 10,000 images, those could be obtained and divided into 4 classes (2500 images each): abnormal in SA+AV node, abnormal in bundle branch, abnormal in Purkinje fiber, and normal condition. Furthermore, we used Transfer learning Xception and Transfer learning MobileNetV2 for training and result comparison. In conclusion, experimental results showed that Transfer learning Xception provided an accuracy of 98.89% and an F1-Score of 97.82%. On the other hand, Transfer learning MobileNetV2 provided an accuracy of 98.44% and an F1-Score of 96.3%. Finally, we took Transfer learning Xception as the best-performance model. This model proceeded to the object detection model and the results of this model showed a detection accuracy between 97% - 98%. The developed model can be used in areas where traditional EKG machines are still used and can help determine the type of abnormality in each type shown EKG and can save time in diagnosis.