| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.65 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A síncope é frequentemente referida como a perda súbita de consciência, associada à
incapacidade de manter o tónus postural, com uma recuperação rápida e espontânea. Mesmo que não
haja risco de mortalidade, é potencialmente perigoso em pessoas com profissões de alto risco e em
pacientes mais velhos que não apresentam sintomas de alerta porque a perda transitória de consciência
pode levar a quedas e lesões. O processamento avançado de sinais em combinação com a integração de
sistemas inovadores construídos com sensores wearable abriram caminho para o desenvolvimento de
novas abordagens para a detecção precoce de síncope iminente. O desenvolvimento destes sensores e a
análise de dados de pré-síncope podem permitir o estabelecimento de um sistema preditivo capaz de
alertar o paciente antes do desmaio, prevenindo muitos riscos e fazendo com que a sua qualidade de
vida aumente substancialmente. A síncope vasovagal é tipicamente caracterizada por desencadeadores
precipitantes e sintomas prodrómicos devido à ativação do sistema autónomo. Normalmente é realizado
um teste de inclinação para induzir a síncope, sendo que este serve como um método amplamente
utilizado para avaliar a suscetibilidade do indivíduo à síncope neuromediada. Este teste também pode
ser um método valioso para educar os indivíduos suscetíveis sobre o diagnóstico de uma condição sem
risco de vida e também pode ajudar esses pacientes a identificar sintomas de alerta.
O teste geralmente começa com um período de repouso em posição supina, seguido por um
período de inclinação passiva (sem o uso de reforços farmacológicos) e depois um período de inclinação
ativa (com o uso de reforços farmacológicos). As respostas ao teste podem ser classificadas em três
tipos: tipo 1 ou misto; tipo 2A ou cardioinibitório sem assistolia; tipo 2B ou cardioinibitório com
assistolia; e tipo 3 ou vasodepressor. O teste é terminado quando o protocolo é concluído na ausência
de sintomas ou quando ocorre síncope. Este é positivo com a indução de síncope ou pré-síncope, e
negativo se o paciente completa o protocolo sem nenhumas alterações hemodinâmicas características.
Com o objetivo de se obter um diagnóstico mais preciso de pacientes com perda transitória de
consciência e distinguir síncope, crises epilépticas e eventos psicogénicos, o teste de inclinação com a
adição de registo de eletroencefalograma (EEG) é proposto para fornecer informações clínicas
adicionais. O EEG permite a avaliação dinâmica da função neurológica e fornece um marcador objetivo
da disfunção cerebral durante a hipoperfusão cerebral que acompanha a síncope. Consequentemente, a
introdução da monitorização de EEG durante o teste pode melhorar significativamente a compreensão
dos eventos cerebrais que ocorrem durante a síncope vasovagal induzida pela inclinação e o potencial
perigo que pode ser causado ao paciente por este procedimento diagnóstico. Durante um episódio
sincopal, foi observado em vários estudos que geralmente o EEG inicialmente mostra um fundo
difusamente retardado e, posteriormente, atividade delta de alta voltagem, geralmente após um intervalo
de cerca de 10 segundos e seguido por vezes de um silêncio eletrocerebral transitório se a hipoperfusão
persistir. Desta forma, o objetivo desta dissertação é implementar um algoritmo que detete a sequência
de alterações no sinal EEG, quando os pacientes estão em iminência de síncope ou apresentam sintomas.
Uma vez que a medição do electrocardiograma e da pressão arterial apenas são capazes de
diagnosticar, mas não podem prever a iminência da síncope, foi proposto o uso de uma banda que mede
a atividade cerebral através de encefalografia, tendo o BITalino como o seu núcleo funcional. Os dados
foram recolhidos por mim com esta banda EEG móvel enquanto os pacientes realizavam o teste no
Hospital de Santa Marta. Foram recolhidos dados de 36 pacientes, dos quais 20 tiveram um teste positivo
e 16 um teste negativo, sendo que, entre os negativos, 8 apresentaram sintomas durante o teste e 8 não. Devido à semelhança nas características do sinal entre os casos positivos e os casos negativos de
pacientes que apresentaram sintomas durante o teste, foi feita uma divisão entre esses 28 casos e os 8
casos de pacientes que não apresentaram qualquer tipo de sintomas durante o teste.
Posteriormente, os dados foram processados, filtrados e segmentados em intervalos de interesse,
sendo estes: 1 minuto durante a fase de repouso em posição supina, o primeiro minuto da fase ativa e 1
minuto, 30 segundos e 10 segundos respetivamente antes da síncope, ou do final da fase ativa no caso
de um teste negativo. Os dados foram analisados em termos de amplitude e frequência. Em termos de
amplitude, foram analisadas apenas as suas variações de intervalo para intervalo, com um grande
aumento da amplitude do sinal observado durante a síncope, que não foi observado nos casos negativos
sem sintomas. Para a frequência, foi feito o espectro de potência do sinal usando o método de Welch em
cada intervalo, sendo este então dividido de acordo com os intervalos do espectro de frequência
conhecidos na literatura. Notou-se um grande aumento na magnitude das ondas delta do intervalo de
repouso para os intervalos anteriores à síncope, o que não ocorreu nos casos negativos sem sintomas ou
nos outros intervalos de frequência.
Concluiu-se então que o que seria mais confiável e mais adequado ao objetivo pretendido, seria
desenvolver o classificador de predição baseado na ocorrência de sintomas e não na síncope, uma vez
que só assim o paciente pode ser avisado com antecendência e podem ser tomadas as devidas
precauções. Os valores das frequências delta de certa forma refletem o que acontece com os resultados
da amplitude. Foi verificado um aumento destes valores em pacientes sintomáticos do intervalo de
repouso até o primeiro minuto da fase ativa, e um aumento ainda maior para os intervalos anteriores ao
término da fase ativa.
As características do sinal EEG durante um episódio sincopal estudado por outras investigações
foram verificadas nos resultados obtidos, onde essas variações foram utilizadas para desenvolver um
modelo classificador baseado na categorização desses dados. Um método de classificação dos resultados
em pacientes sintomáticos e não sintomáticos teve que ser desenvolvido para que os resultados obtidos
pudessem ser colocados em prática e testados num contexto prático para confirmar a sua veracidade.
Desta forma, os dados de cada paciente individual foram classificados de acordo com o seu valor, tanto
na análise realizada em termos de amplitude como na de frequência. A classificação foi feita através de
um classificador de vetores de suporte, dividindo os dados num conjunto de treino e num conjunto de
teste. Esta metodologia foi aplicada na variação dos valores de amplitude e frequência entre o intervalo
de repouso e o intervalo de 1 minuto antes do fim da fase ativa, bem como entre o intervalo de repouso
e o intervalo de 30 segundos antes do fim da fase ativa. Foi desenvolvido ainda um hiperplano para as
duas variáveis em função uma da outra de modo a separar as classes e classificar o conjunto de teste.
Infere-se que o modelo possui alguma confiabilidade, tendo conseguido categorizar
corretamente todas as observações. Para o modelo com os resultados das variações entre o intervalo de
repouso e 1 minuto antes do término da fase ativa, a sua precisão de teste teve um valor médio de 100%,
sendo que para o intervalo de 30 segundos teve uma precisão de teste com uma média de 93,8%. Pensase que com um número maior de amostras este modelo poderá ser implementado num contexto prático.
Assim, conclui-se que o estudo realizado apresenta resultados muito promissores e inovadores tanto no
que diz respeito ao padrão encontrado nos dados recolhidos, como na sua classificação, apesar de todas
as limitações inerentes ao teste de inclinação e ao estudo em questão. O trabalho a ser feito no futuro
passa justamente por implementar este algoritmo, num sistema móvel de EEG conectado ao telemóvel
do paciente, que o avise com tempo suficiente para que a sua qualidade de vida aumente
substancialmente.
Syncope is often referred as the sudden loss of consciousness, associated with inability to maintain postural tone, with a spontaneous and rapid recovery. Even though it is not associated with an increased risk of mortality, it is potentially dangerous in those with high-risk occupations and in older patients who lack warning symptoms because the transient loss of consciousness (TLOC) may lead to falls and injury. Vasovagal syncope (VVS) is typically characterized by precipitating triggers and prodromal symptoms. Usually is performed an head-up tilt test (HUTT) to induce syncope, serving as a widely used modality for diagnosing this condition. The HUTT usually starts with a period of supine rest, followed by a period of passive tilt (without the use of pharmacological enhancement) and then a period of active tilt (with the use of pharmacological enhancement). The test is ended when the protocol is completed in the absence of symptoms, or there is occurrence of syncope. The HUTT is positive with the induction of syncope or presyncope and is negative if the patient completes the protocol with no characteristic hemodynamic changes. It is thought that the introduction of EEG monitoring during head-up tilt testing could significantly improve the understanding of the cerebral events occurring during tilt-induced vasovagal syncope since the EEG provides an objective marker of brain dysfunction during the cerebral hypoperfusion that accompanies syncope. During a syncopal episode, the EEG generally is said to show a diffusely slowed background initially and then high-voltage delta activity. This way, the objective of this dissertation is to implement an algorithm that detects the sequence of changes in the EEG signal, when patients are in imminent syncope, or show symptoms of it. The data was measured by me with a wearable EEG band that has BITalino as its functional core while patients performed the test. Data were taken from 36 patients, 20 of whom had a positive test and 16 had a negative test, being that among the negative, 8 had symptoms during the test and 8 did not. Due to the similarity in the signal characteristics between the positive cases and the negative cases of patients who had symptoms during the test, a division was made between these 28 cases and the 8 cases of patients who did not experience any symptoms. The data was processed, filtered and segmented in intervals of interest, these being, 1 minute during the supine rest phase, the first minute of the active phase and 1 minute, 30 seconds and 10 seconds respectively before syncope, or the end of active phase in the case of a negative test. Data were analyzed in terms of amplitude, where a large increase in signal amplitude was observed during syncope, which was not observed in negative cases without symptoms. In frequency, the power spectrum of the signal in each interval was made, and this was then divided according to the frequency spectrum intervals known in the literature. A large increase in the power of delta waves from the supine rest interval to the intervals before syncope was noted, which did not occur in the negative cases without symptoms or in with the other frequency spectra. It was concluded that what would be more reliable and more adjusted to the intended objective, would be to develop the prediction classifier based on symptoms and not on syncope. It was verified an increase in values of both amplitude and frequency in symptomatic patients from the supine rest interval to the first minute of the active phase, and a further increase to the intervals before the end of the active phase. The characteristics in the EEG signal during a syncopal episode studied by other investigations were verified in the results obtained, where these variations were used to develop a classifier model to categorize the results in symptomatic patients and non-symptomatic patients. The classification was done through a train support vector machine (SVM) classifier for the variation in amplitude and frequency values between the supine rest and the 1-minute interval and the 30-seconds interval before syncope. It is inferred that the model has some reliability, having managed to correctly categorize all observations. For the model with the results of variations between the supine rest interval and 1 minute before the end of the active phase, its accuracy had a mean value of 100% and for the 30-second interval had a test accuracy with a mean of 93.8%. It is thought that with a larger number of samples, it can be implemented in a practical context. Thus, it is concluded that the study carried out has very promising and innovative results in terms of the pattern found, as well as in its classification.
Syncope is often referred as the sudden loss of consciousness, associated with inability to maintain postural tone, with a spontaneous and rapid recovery. Even though it is not associated with an increased risk of mortality, it is potentially dangerous in those with high-risk occupations and in older patients who lack warning symptoms because the transient loss of consciousness (TLOC) may lead to falls and injury. Vasovagal syncope (VVS) is typically characterized by precipitating triggers and prodromal symptoms. Usually is performed an head-up tilt test (HUTT) to induce syncope, serving as a widely used modality for diagnosing this condition. The HUTT usually starts with a period of supine rest, followed by a period of passive tilt (without the use of pharmacological enhancement) and then a period of active tilt (with the use of pharmacological enhancement). The test is ended when the protocol is completed in the absence of symptoms, or there is occurrence of syncope. The HUTT is positive with the induction of syncope or presyncope and is negative if the patient completes the protocol with no characteristic hemodynamic changes. It is thought that the introduction of EEG monitoring during head-up tilt testing could significantly improve the understanding of the cerebral events occurring during tilt-induced vasovagal syncope since the EEG provides an objective marker of brain dysfunction during the cerebral hypoperfusion that accompanies syncope. During a syncopal episode, the EEG generally is said to show a diffusely slowed background initially and then high-voltage delta activity. This way, the objective of this dissertation is to implement an algorithm that detects the sequence of changes in the EEG signal, when patients are in imminent syncope, or show symptoms of it. The data was measured by me with a wearable EEG band that has BITalino as its functional core while patients performed the test. Data were taken from 36 patients, 20 of whom had a positive test and 16 had a negative test, being that among the negative, 8 had symptoms during the test and 8 did not. Due to the similarity in the signal characteristics between the positive cases and the negative cases of patients who had symptoms during the test, a division was made between these 28 cases and the 8 cases of patients who did not experience any symptoms. The data was processed, filtered and segmented in intervals of interest, these being, 1 minute during the supine rest phase, the first minute of the active phase and 1 minute, 30 seconds and 10 seconds respectively before syncope, or the end of active phase in the case of a negative test. Data were analyzed in terms of amplitude, where a large increase in signal amplitude was observed during syncope, which was not observed in negative cases without symptoms. In frequency, the power spectrum of the signal in each interval was made, and this was then divided according to the frequency spectrum intervals known in the literature. A large increase in the power of delta waves from the supine rest interval to the intervals before syncope was noted, which did not occur in the negative cases without symptoms or in with the other frequency spectra. It was concluded that what would be more reliable and more adjusted to the intended objective, would be to develop the prediction classifier based on symptoms and not on syncope. It was verified an increase in values of both amplitude and frequency in symptomatic patients from the supine rest interval to the first minute of the active phase, and a further increase to the intervals before the end of the active phase. The characteristics in the EEG signal during a syncopal episode studied by other investigations were verified in the results obtained, where these variations were used to develop a classifier model to categorize the results in symptomatic patients and non-symptomatic patients. The classification was done through a train support vector machine (SVM) classifier for the variation in amplitude and frequency values between the supine rest and the 1-minute interval and the 30-seconds interval before syncope. It is inferred that the model has some reliability, having managed to correctly categorize all observations. For the model with the results of variations between the supine rest interval and 1 minute before the end of the active phase, its accuracy had a mean value of 100% and for the 30-second interval had a test accuracy with a mean of 93.8%. It is thought that with a larger number of samples, it can be implemented in a practical context. Thus, it is concluded that the study carried out has very promising and innovative results in terms of the pattern found, as well as in its classification.
Descrição
Tese de Mestrado Integrado, Engenharia Biomédica e Biofísica (Engenharia Clínica e Instrumentação Médica), 2022, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Síncope vasovagal Teste de inclinação Banda EEG móvel Densidade espetral Classificador de vetores de suporte Teses de mestrado - 2022
