Electrocardiogram based heart disease diagnosis using artificial intelligence
Heart diseases have been the major cause of deaths in the world according to a recent study. The main tool that is widely used to understand the cardiac condition is an Electrocardiogram (ECG). Normal and abnormal cardiac function of the human heart can be analyzed through the application of the ECG...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2015
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/13096/19/Electrocardiogram%20based%20heart%20disease%20diagnosis%20using%20artificial%20intelligence.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Heart diseases have been the major cause of deaths in the world according to a recent study. The main tool that is widely used to understand the cardiac condition is an Electrocardiogram (ECG). Normal and abnormal cardiac function of the human heart can be analyzed through the application of the ECG signal processing and evaluation. Although traditionally the interpretation of these signals remains largely a manual effort, as computing power has increased, so too has the application of computational methods for ECG evaluation and classification. Therefore, it is necessary to have suitable methods for early detection of heart condition of the patient. In addition, the ECG is recorded on a thermal paper, which cannot be stored for a long time, because thermal trace over time tends to erase gradually. However, some hospitals are saving the ECG thermal papers as scanning images in the electronic equipment's (like computers) to maintain medical records, but this method requires too high memory capacity, in addition to use low scanning resolution that gives ECG image unclear when preview. In this thesis, image-processing techniques are developed for the ECG feature extraction and signal regeneration as a digital time series signal. The first step is to use a flatbed scanner to take an image of the ECG signals; during this step the bit depth, the image resolution and output file format are a main concern. There are chances that the image may be blurring or suffer from some chromatography ambiguity due to unclear original image. Therefore, a new method for enhancing the image contrast called Fuzzy Hyperbolic Threshold is proposed with a new membership equation. This method has significant impact on the adjustment of lighting in dark images, clarifies its edges, clarifies their features, and improved image quality. In addition can use this method on different types of medical images, and the simulation results have been a very good when compared with the other methods that can used in the image contrast enhancement. Accordingly, the first step is an image segmentation method using proposed thresholding algorithms has been used to locate objects and boundaries of the ECG signal and background grid lines in the ECG images. This technique is used to transform the data of the ECG signal recorded on paper to a digital time series database. The ECG signal is usually infected with various kinds of noise such as Baseline Wander (BW) noise, Motion Artifacts (EA) noise, Muscle Artefacts (MA) noise, and Power Line Interference (PLI) noise. The useful and powerful techniques that use to analyze such this type of signals are an adaptive Wavelet filter. The distortion in the S-T segment of ECG signal can be minimized by applying a new technique, which is, amalgamates the adaptive filter and hybrid soft computing technique known as Discrete Wavelet Transform (DWT) for BW noise removal. After noise removal, the data from the ECG is to be acquired; for this purpose a method is devised based on DWT. Third stage is the features extraction; proposed a special domain based on DWT to extract diagnostic information from the ECG signal. Symlet transform (one of the wavelet transform families) was used for acquiring the desired data from ECG signals and achieved 99.50% productivity and 99.87% sensitivity. Finally, five different types of the ECG diseases are identified using various artificial intelligence types like hybrid RBFNN and hybrid PSO-RBFNN. The optimal neural network model (PSO-RBFNN) has 13 input nodes, 40 hidden nodes, and 5 output nodes signifying Angina, RBBB, MI, Normal and LBBB. The classification performance was carried out with 99.59% for specificity and 98.37% for sensitivity. |
|---|