Classifying Normal and Abnormal Heartbeats From a Noisy ECGOutlineFiltering – High-PassFiltering – Band-PassBeat DetectionMLP Beat ClassificationSlide 7SVM Beat ClassificationSlide 9ConclusionQuestions?Classifying Normal Classifying Normal and Abnormal and Abnormal Heartbeats From a Heartbeats From a Noisy ECGNoisy ECGEric PetersonEric PetersonECE 539ECE 539OutlineOutlineFiltering – Some BasicsFiltering – Some BasicsBeat Detection – FailedBeat Detection – FailedMLP Beat Classification – Works…MLP Beat Classification – Works…SometimesSometimesSVM Beat Classification – Similar SVM Beat Classification – Similar ResultsResultsConclusion – More Pre-Processing Conclusion – More Pre-Processing NeededNeededFiltering – High-PassFiltering – High-Pass0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16-80-70-60-50-40-30-20-10010Frequency (kHz)Magnitude (dB)Magnitude Response (dB)Filtering – Band-PassFiltering – Band-Pass0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16-120-100-80-60-40-200Frequency (kHz)Magnitude (dB)Magnitude Response (dB)Beat DetectionBeat DetectionSupplied the Filtered SignalSupplied the Filtered SignalOverwhelmed the ANNOverwhelmed the ANNSNR does not matterSNR does not matterFAILURE!!!FAILURE!!!Pan-TompkinsPan-TompkinsOverwhelmed againOverwhelmed againMay not actually be linearly seperableMay not actually be linearly seperableModifications requredModifications requredMLP Beat ClassificationMLP Beat ClassificationUsed annotations to focus on beats onlyUsed annotations to focus on beats onlyAnnotations of either normal or Annotations of either normal or abnormal beatsabnormal beatsAttempted many parameter variationsAttempted many parameter variationsBest classification rate: 95.8824%Best classification rate: 95.8824%Confusion Matrix: 159Confusion Matrix: 15922 8844Results were dominated by the normal beatsResults were dominated by the normal beatsFailed with a SNR<24dBFailed with a SNR<24dBMLP Beat ClassificationMLP Beat ClassificationInputs Learning Rate Momentum Hidden Layers Classification Rate Confusion Matrix2 0.001 1 2 95.8824 159 2 BEST8 42 0.01 0.001 2 95.2941 159 17 32 0.01 0.01 2 95.2941 159 17 32 0.01 0.1 2 95.2941 159 17 32 0.01 1 2 95.2941 159 27 32 0.1 0.5 3 95.2941 159 17 310 0.1 0.5 7 95.8824 160 0 BEST7 350 0.01 0.5 5 95.2941 160 08 2SVM Beat ClassificationSVM Beat ClassificationRBF kernel did not RBF kernel did not workworkSimilar results to MLPSimilar results to MLPStill seems dominated Still seems dominated by the normal beatsby the normal beatsFailed at <24dB SNRFailed at <24dB SNRInputs Kernel Type Accuracy Confusion Matrix2 Polynomial 93.53% 158 29 15 Polynomial 92.94% 155 57 310 Sigmoid 94.71% 156 45 550 Sigmoid 93.53% 158 29 1SVM Beat ClassificationSVM Beat ClassificationConclusionConclusionMore Pre-Processing is needed!!!More Pre-Processing is needed!!!Possibility of better filtering?Possibility of better filtering?Further analysis of the signalFurther analysis of the signalFeed the neural nets with important valuesFeed the neural nets with important valuesTemplates were used in many previous Templates were used in many previous paperspapersNot ideal for many types of abnormal beatsNot ideal for many types of abnormal
View Full Document
Unlocking...