| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 3.04 MB | Adobe PDF |
Authors
Abstract(s)
A doença da apneia do sono afeta cerca de 1 milhão de pessoas no mundo. Ela distinguese
em três tipos: Obstrutiva, Central e Mista. A investigação por uma alternativa à Polissonografia como meio de diagnóstico à Apneia Obstrutiva do Sono é uma necessidade, visto
os custos e meios que ela envolve. Nesse sentido a presente dissertação descreve a envolvência do problema e propõe modelo de deteção de episódios de apneia, com recurso
ao aprendizado de máquina (em inglês, Machine Learning).
O modelo foi construído com o uso de diferentes classificadores (SVM, MLP, Adaboost e
RandomForest). Como dados de entrada, foram processados e extraídas caraterísticas do
base de dados ApneiaECG da Physionet. O melhor resultado alcançado foi uma exatidão
(accuracy) de 78,45%, sensibilidade (sensibility) de de 70,26% e especificidade specificity
de 83,49%.
Adicionalmente, foi construída uma aplicação de recolha de dados que utiliza um dispositivo capacitado com elétrodos que permitem a recolha de sinal de eletrocardiograma.
Com estes dados, pretendese no futuro criar uma base de dados e um sistema que ajude
a deteção da apneia.
Sleep apnea disease affects about 1 million people worldwide. It is distinguished into three types: Obstructive, Central and Mixed. The investigation for an alternative to Polysomnography as a means of diagnosing Obstructive Sleep Apnea is a necessity, given the costs and means that it involves. In this sense, the present dissertation describes the surroundings of the problem and proposes a model for the detection of apnea episodes, using machine learning (in English, Machine Learning). The model was built using different classifiers (SVM, MLP, Adaboost and RandomForest). As input data, features were processed and extracted from the ApneaECG database of Physionet. The best result achieved was an accuracy (accuracy) of 78.45%, sensitivity (sensibility) of 70.26% and specificity specificity of 83.49%. Additionally, a data collection application was built that uses a device equipped with electrodes that allow the collection of an electrocardiogram signal. With these data, it is intended in the future to create a database and a system to help detect apnea.
Sleep apnea disease affects about 1 million people worldwide. It is distinguished into three types: Obstructive, Central and Mixed. The investigation for an alternative to Polysomnography as a means of diagnosing Obstructive Sleep Apnea is a necessity, given the costs and means that it involves. In this sense, the present dissertation describes the surroundings of the problem and proposes a model for the detection of apnea episodes, using machine learning (in English, Machine Learning). The model was built using different classifiers (SVM, MLP, Adaboost and RandomForest). As input data, features were processed and extracted from the ApneaECG database of Physionet. The best result achieved was an accuracy (accuracy) of 78.45%, sensitivity (sensibility) of 70.26% and specificity specificity of 83.49%. Additionally, a data collection application was built that uses a device equipped with electrodes that allow the collection of an electrocardiogram signal. With these data, it is intended in the future to create a database and a system to help detect apnea.
Description
Keywords
Support Machine Vetor (Svm) Adaboost Aplicação Móvel Apneia Obstrutiva do Sono (Aos) Classificadores Deteção Diagnóstico EcgDerived Respiration (Edr) Eletrocardiograma (Ecg) Heart Rate Variability Machine Learning MultiLayer Perceptron (Mlp) Randomforest Recolha de Dados