| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 9.23 MB | Adobe PDF |
Authors
Abstract(s)
As doenças de Alzheimer e Parkinson destacam-se como as duas patologias neurodegenerativas com maior incidência no mundo. O impacto sobre milhões de pessoas, aliado à inexistência de uma cura para ambas as patologias e ao seu período inicial assintomático, evidencia a necessidade de um diagnóstico o mais precoce possível, que possibilite um controlo do desenvolvimento das mesmas. De forma a colmatar a falha clínica de diagnóstico na fase inicial do desenvolvimento das patologias, tirou-se partido da natureza não linear inerente ao sinal EEG, em conjunto com a aplicação da Transformada Wavelet Discreta. Esta ferramenta tem demonstrado vantagens na análise das bandas frequenciais do sinal e em conjunto com uma análise não linear do mesmo, poderá fornecer informação adicional que permita maximizar a deteção e diferenciação de ambas as patologias na sua fase inicial. A classificação dos sinais foi realizada através de diversos classificadores como Árvores de Decisão, Support Vector Machines, K-nearest-neighbors e Regressão Logística, num processo de validação cruzada em leave-one-out. A comparação dos grupos de Parkinson e Alzheimer alcançou precisões de 100% em pelo menos um canal nas bandas beta, alfa, teta e delta. As comparações individuais dos grupos Parkinson e Alzheimer face ao grupo de controlo obtiveram em ambos os casos precisões máxi-mas de 100%, em pelo menos um canal na banda delta, com a comparação do grupo de Parkinson com controlo a atingir também 100% num canal da banda gama. Finalmente, a comparação em conjunto dos grupos de controlo atingiu precisão de 90,6% na bandateta.
Alzheimer’s and Parkinson’s diseases stand out as the two most common neurodegenerative pathologies in the world. The impact on millions of people, combined with the lack of a cure for both diseases and their asymptomatic initial period, highlights the need for diagnosis at the earliest possible stage to control their development. In order to remedy the clinical failure of diagnosis in the initial phase of their development, the non-linear nature inherent to the EEG signal was used together with the application of the Discrete Wavelet Transform. This tool has demonstrated advantages in the analysis of the signal frequency bands and together with a non-linear analysis of the signal can provide additional information to maximise the detection and differentiation of both pathologies in their initial phase. The classification of the signals has been performed through several classifiers such as Decision Trees, Support Vector Machines, K-nearest-neighbors and Logistic Regression, in a process of leave-one-out cross-validation. The comparison of Parkinson’s and Alzheimer’s groups reached accuracies of 100% on at least one channel in the Beta, Alfa, Teta and Delta bands. Individual comparisons of Parkinson’s and Alzheimer’s groups against the control group achieved in both cases maximum accuracies of 100% on at least one channel in the delta band, with the comparison of the Parkinson’s group against control also reaching 100% on a gamma band channel. Finally, the comparison of Parkinson’s and Alzheimer’s groups together against the control group achieved an accuracy of 90.6% in theta band.
Alzheimer’s and Parkinson’s diseases stand out as the two most common neurodegenerative pathologies in the world. The impact on millions of people, combined with the lack of a cure for both diseases and their asymptomatic initial period, highlights the need for diagnosis at the earliest possible stage to control their development. In order to remedy the clinical failure of diagnosis in the initial phase of their development, the non-linear nature inherent to the EEG signal was used together with the application of the Discrete Wavelet Transform. This tool has demonstrated advantages in the analysis of the signal frequency bands and together with a non-linear analysis of the signal can provide additional information to maximise the detection and differentiation of both pathologies in their initial phase. The classification of the signals has been performed through several classifiers such as Decision Trees, Support Vector Machines, K-nearest-neighbors and Logistic Regression, in a process of leave-one-out cross-validation. The comparison of Parkinson’s and Alzheimer’s groups reached accuracies of 100% on at least one channel in the Beta, Alfa, Teta and Delta bands. Individual comparisons of Parkinson’s and Alzheimer’s groups against the control group achieved in both cases maximum accuracies of 100% on at least one channel in the delta band, with the comparison of the Parkinson’s group against control also reaching 100% on a gamma band channel. Finally, the comparison of Parkinson’s and Alzheimer’s groups together against the control group achieved an accuracy of 90.6% in theta band.
Description
Keywords
Parkinson Alzheimer EEG Análise não Linear Transformada wavelet Mild cognitive impairment Non-linear analysis Wavelet transform Mild cognitive impairment