| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.82 MB | Adobe PDF |
Advisor(s)
Abstract(s)
Com o aumento exponencial dos casos de doenças cardiovasculares, a idealização de um algoritmo
que possibilita a distinção de doenças é um grande aliado no diagnóstico. O bloqueio do ramo direito, apesar de ser uma doença que pode nunca vir a apresentar sintomas, é um excelente indicador de possíveis doenças cardiovasculares futuras. De forma a detetar o aparecimento do BBB nas suas fases iniciais, neste trabalho aplicou-se aos sinais ECG a Transformada Wavelet Discreta, o que permitiu extrair características na forma de energia, entropia e coerência de três níveis diferentes da decomposição do sinal. A discriminação dos sinais foi realizada através da CNN no processo de validação cruzada 30-fold.
A precisão na comparação entre o BBB e as outras doenças, presentes na mesma base de dados,
cifrou-se entre 98,90% e 100% utilizando pequenas porções de sinal com pares de entrada/saída para
as CNNs. No caso da medida energética as CNNs conseguiram precisões entre 91,14% e 66,51%, para
a entropia, 91,68% e 64,31% e utilizando a coerência obteve-se uma precisão máxima de 90,83%.
With the exponential growing up in the number of cases for cardiovascular diseases, the idealization of an algorithm that can distinguish pathologies is a great ally in diagnosis. The Right Bundle Branch Block, even though is a disease that can never present symptoms, it is an excellent indicator for future cardiovascular diseases. In order to detect the apppearence of the BBB in the early stages, in this work the Discrete Wavelet Transform was applied to the ECG signals, which allowed to extract characteristics such as energy, entropy and coherence from three different levels of signal decomposition. The signal discrimination was performed through CNN in the 30-fold cross-validation process. The comparison accuracy between BBB and other diseases, present in the database, ranged from 98,90% and 100% using small portions of signal as input/output pairs fotr the CNN. In the case of the energy measurement, CNN provided an accuracy between 91,14% and 66,51%, for the entropy, 91,68% and 64,31% and using coherences, a maximum accuracy of 90,83%.
With the exponential growing up in the number of cases for cardiovascular diseases, the idealization of an algorithm that can distinguish pathologies is a great ally in diagnosis. The Right Bundle Branch Block, even though is a disease that can never present symptoms, it is an excellent indicator for future cardiovascular diseases. In order to detect the apppearence of the BBB in the early stages, in this work the Discrete Wavelet Transform was applied to the ECG signals, which allowed to extract characteristics such as energy, entropy and coherence from three different levels of signal decomposition. The signal discrimination was performed through CNN in the 30-fold cross-validation process. The comparison accuracy between BBB and other diseases, present in the database, ranged from 98,90% and 100% using small portions of signal as input/output pairs fotr the CNN. In the case of the energy measurement, CNN provided an accuracy between 91,14% and 66,51%, for the entropy, 91,68% and 64,31% and using coherences, a maximum accuracy of 90,83%.
Description
Keywords
ECG Bloqueio do ramo direito Transformada de wavelet discreta Convolutional neural network Validação cruzada Right bundle branch block Discrete wavelet transform Convolutional neural network Cross-validation