Classifying Normal and Abnormal Heartbeats From a Noisy
Classifying Normal and Abnormal Heartbeats From a Noisy ECG Eric Peterson ECE 539
Outline Filtering – Some Basics n Beat Detection – Failed n MLP Beat Classification – Works…Sometimes n SVM Beat Classification – Similar Results n Conclusion – More Pre-Processing Needed n
Filtering – High-Pass
Filtering – Band-Pass
Beat Detection n Supplied the Filtered Signal Overwhelmed the ANN n SNR does not matter n FAILURE!!! n n Pan-Tompkins Overwhelmed again n May not actually be linearly seperable n n Modifications requred
MLP Beat Classification Used annotations to focus on beats only n Annotations of either normal or abnormal beats n Attempted many parameter variations n Best classification rate: 95. 8824% n Confusion Matrix: 159 2 8 4 n Results were dominated by the normal beats n n Failed with a SNR<24 d. B
MLP Beat Classification
SVM Beat Classification n n RBF kernel did not work Similar results to MLP Still seems dominated by the normal beats Failed at <24 d. B SNR
SVM Beat Classification
Conclusion n More Pre-Processing is needed!!! Possibility of better filtering? n Further analysis of the signal n n Feed the neural nets with important values n Templates were used in many previous papers n Not ideal for many types of abnormal beats
Questions? http: //www. metamemes. com
- Slides: 11