Loading...
Please wait, while we are loading the content...
Similar Documents
Universidade Federal De Santa Catarina Programa De Pós-graduação Em Engenharia Elétrica
| Content Provider | Semantic Scholar |
|---|---|
| Author | Zardo, Fernanda |
| Copyright Year | 2014 |
| Abstract | This research presents a methodology for the identification of epileptiform paroxysms in EEG signals based on Wavelet scalogram that maps the signal in time and scale using a Wavelet function. It was used 65 Wavelet functions of families: Daubechies, Biorthogonal, Symlets, Reverse Biorthogonal and Coiflets. After feature extraction via scalograms it was designed a Multi-Layer Perceptron (MLP) artificial neural network to identify the epileptiform events (spikes and sharp waves). Two banks of signals were used: EEG-Bank-A and EEG-Bank-B which are totally different and they will help to test the proposed methodology. It was proposed two ways for the training stage: using the full dyadic scalogram or the dyadic scales more strongly related to epileptiform activity, the dyadic scales: 25, 26, 27 and 28. The purpose is to decrease high redundancy of information of the CWT also reducing the high computational cost. It was trained 260 neural networks using the same vector of initial weights. The tests were performed using a cross-data technique (between the banks), generating the following indicators of performance: sensitivity, specificity, positive and negative predictive values, prevalence, maximum efficiency and area under the ROC curve (AUC). The Wavelet functions were evaluated based on the AUC x EFI product. For EEG-Bank-A the functions bior3.7, bior3.9 and rbio1.5 were chosen obtaining the indicators of performance: sensitivity of 78.21%, specificity of 94.53%, positive predictive value of 89.97%, negative predictive value of 87.33%, prevalence of 38.62%, maximum efficiency of system of 88.22% and AUC of 0.9617. For EEG-Bank-B were chosen rbio1.5, rbio1.3 and coif1 obtaining the indicators: sensitivity of 89.03%, specificity of 89.33%, positive predictive value of 85.40%, negative predictive value of 92.07%, prevalence of 41.21%, maximum efficiency of 89.20% and AUC of 0.9461. The rbio1.5 function provides high indicators of performance for both banks. In general, all Wavelet functions are useful for the identification of epileptiform paroxysms, even though the function daub10 to daub15 reached AUC x EFI indicators smaller than 75% that was considered a low value. Finally, the processing time of the proposed system was 2.5 seconds. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://ppgeel.posgrad.ufsc.br/files/2018/05/Edital-Mestrado-2018-2.pdf |
| Alternate Webpage(s) | http://lcs.ufsc.br/files/2018/03/326662.pdf |
| Alternate Webpage(s) | https://repositorio.ufsc.br/xmlui/bitstream/handle/123456789/100492/307754.pdf?isAllowed=y&sequence=1 |
| Alternate Webpage(s) | https://repositorio.ufsc.br/bitstream/handle/123456789/158375/336603.pdf?isAllowed=y&sequence=1 |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
