Browsing by Author "Borghi, Pedro Henrique"
Now showing 1 - 7 of 7
Results Per Page
Sort Options
- Android-based ECG monitoring system for atrial fibrillation detection using a BITalino® ECG sensorPublication . Lazaretti, Gabriel Saatkamp; Teixeira, João Paulo; Kuhn, Eduardo Vinicius; Borghi, Pedro HenriqueCardiac arrhythmias are disorders that affect the rate and/or rhythm of the heartbeats. The diagnosis of most arrhythmias is made through the analysis of the electrocardiogram (ECG), which consists of a graphical representation of the electrical activity of the heart. Atrial fibrillation (AF) is the most present type of arrhythmia in the world population. In this context, this work deals with the implementation of a system for automatic analysis of ECG signals aiming to identify AF episodes. The system consists of a signal acquisition step performed by an ECG sensor connected to an acquisition platform. The acquired signal is transmitted via bluetooth to a smartphone with Android™ operating system. The signal processing is carried out through an application developed using the IDE Android™ Studio. When assessed over signals from the MIT-BIH Atrial Fibrillation database, the R-wave peak detection algorithm showed mean values of sensitivity and positive predictivity of 98.99% and 95.95%, respectively. The classification model used is based on a long short-term memory (LSTM) neural network and had an average accuracy of 94.94% for identifying AF episodes.
- Atrial fibrillation classification based on MLP networks by extracting Jitter and Shimmer parametersPublication . Borghi, Pedro Henrique; Borges, Renata Coelho; Teixeira, João PauloAtrial fibrillation (AF) is the most common cardiac anomaly and one that potentially threatens human life. Due to its relation to a variation in cardiac rhythm during indeterminate periods, long-term observations are necessary for its diagnosis. With the increase in data volume, fatigue and the complexity of long-term features make analysis an increasingly impractical process. Most medical diagnostic aid systems based on machine learning, are designed to automatically detect, classify or predict certain behaviors. In this work, using the PhysioNet MIT-BIH Atrial Fibrillation database, a system based on MLP artificial neural network is proposed to differentiate, between AF and non-AF, segments and ECG’s features, obtaining average accuracy of 80.67% in test set, for the 10-fold cross-validation method. As a highlight, the extraction of jitter and shimmer parameters from ECG windows is presented to compose the network input sets, indicating a slight improvement in the model's performance. Added to these, Shannon's and logarithmic energy entropies are determined, also indicating an improvement in performance related to the use of fewer features.
- Automatic Speech Recognition for Portuguese: A Comparative StudyPublication . Borghi, Pedro Henrique; Teixeira, João Paulo; Freitas, Diamantino RuiThis paper provides some comparisons of Automatic Speech Recognition (ASR) services for Portuguese that were developed in the scope of the Safe Cities project. ASR technology has enabled bi-directional voice-driven interfaces, and its demand in Portuguese is evident due to the language’s global prominence. However, the transcription process has complexities, and a high accuracy depends on the ability of capturing speech variability and language intricacies, while being rigorous in terms of semantics. The study first describes ASR services/models by Google, Microsoft, Amazon, IBM, and Voice Interaction regarding their main features. To compare them, three tests were proposed. Test A uses a small dataset with six audio recordings to evaluate in terms of word hit rate the accuracy of online services, with IBM outperforming others (pt-BR: 93.33%). Tests B and C utilize theMozilla Common Voice database filtered by a keywords’ set to compare online and offline models for Brazilian and European Portuguese regarding accuracy (Ratcliff-Obershelp algorithm), Word Error Rate, Match Error Rate, Word Information Loss, Character Error Rate and Response-Request Ratio. Test B highlights the higher accuracy of Google Cloud (pt-PT: 94.90%) and Azure (pt-BR: 98.11%). Test C showcases the potential of Voice Interaction’s real-time application despite its lower accuracy (pt-PT: 78.81%). The tests were carried out using a framework developed using Python 3.x on a Raspberry Pi 4 model B with a server desktop and the REST APIs from the companies’ repositories.
- Brief review on electrocardiogram analysis and classification techniques with machine learning approachesPublication . Borghi, Pedro HenriqueElectrocardiogram captures the electrical activity of the heart. The signal obtained can be used for various purposes such as emotion recognition, heart rate measuring and the main one, cardiac disease diagnosis. But ECG analysis and classification require experienced specialists once it presents high variability and suffers interferences from noises and artefacts. With the increase of data amount on long term records, it might lead to long term dependencies and the process become exhaustive and error prone. Automated systems associated with signal processing techniques aim to help on these tasks by improving the quality of data, extracting meaningful features, selecting the most suitable and training machine learning models to capture and generalize its behaviour. This review brings a brief stage sense of how data flows into these approaches and somewhat techniques are most used. It ends by presenting some of the countless applications that can be found in the research community.
- Classificação de episódios de fibrilação atrial por análise do ECG com redes neuronais artificiais MLP e LSTMPublication . Borghi, Pedro Henrique; Teixeira, João Paulo; Borges, Renata CoelhoA fibrilação atrial (AF) é uma doença cardíaca que afeta aproximadamente 1% da população mundial, sendo a anomalia cardíaca mais comum. Apesar de não ser uma causa direta de morte, frequentemente está associada ou gera outros problemas que ameaçam a vida humana, como o derrame e a doença da artéria coronária. As principais características da AF são: a alta variação do ritmo cardíaco, o enfraquecimento ou desaparecimento da contração atrial e a ocorrência de irregularidades nas atividades dos ventrículos. O diagnóstico da AF é realizado por um médico especialista, principalmente através da inspeção visual de gravações de eletrocardiograma (ECG) de longo termo. Tais gravações podem chegar a várias horas, e são necessárias pois a AF pode ocorrer a qualquer momento do dia. Dessa forma surgem os problemas quanto ao grande volume de dados e as dependências de longo termo. Além disso, as particularidades e as variabilidades dos padrões de deformação de cada sujeito fazem com que o problema esteja também relacionado com a experiência do cardiologista. Assim, a proposta de um sistema computacional de auxílio ao diagnóstico médico baseado em inteligência artificial se torna muito interessante, uma vez que não sofre com a fadiga e é fortemente indicado para lidar com dados em grande quantidade e com alta variabilidade. Portanto, neste trabalho foi proposta a exploração de modelos de aprendizagem de máquina para análise e classificação de sinais ECG de longo termo, para auxiliar no diagnóstico da AF. Os modelos foram baseados em redes neuronais artificiais do tipo Multi-Layer Perceptron (MLP) e Long Short-Term Memory (LSTM). Utilizam-se os sinais da base de dados MIT-BIH Atrial Fibrillation, sem remoção de ruído, tendências ou artefatos, numa etapa de extração de características temporais, morfológicas, estatísticas e em tempo-frequência sobre segmentos de contexto variável (duração em segundos ou contagem de intervalos entre picos R). As características do sinal ECG utilizadas, foram: duração dos intervalos R-R (RRi) consecutivos, perturbação Jitter, perturbação Shimmer, entropias de Shannon e energia logarítmica, frequências instantâneas, entropia espectral e transformada Scattering. Sobre estes atributos foram aplicadas diferentes estratégias de normalização por Z-score e valor máximo absoluto, de forma a normalizar os indicadores de acordo com o contexto do sujeito ou local do segmento. Após a exploração de várias combinações destas características e dos parâmetros das redes MLP, obteve-se uma acurácia de classificação para a metodologia 10-fold cross-validation de 80,67%. Entretanto, notou-se que as marcações do pico das ondas R advindas da base de dados eram imprecisas. Dessa forma, desenvolveu-se um algoritmo de detecção do pico das ondas R baseado na combinação entre a derivada do sinal, a energia de Shannon e a transformada de Hilbert, resultado em uma acurácia de marcação dos picos R de 98,95%. A partir das novas marcações, determinou-se todas as características e em seguida foram exploradas diversas estruturas de redes neuronais MLP e LSTM, sendo que os melhores resultados em acurácia/exatidão para estas arquiteturas foram, respectivamente, 91,96% e 98,17%. Em todos os testes, a MLP demonstrou melhora de desempenho à medida que mais características foram sendo agregadas nos conjuntos de dados. A LSTM por outro lado, obteve os melhores resultados quando foram combinados 60 RRi e as respectivas entropias das ondas P, T e U.
- A COVID-19 time series forecasting model based on MLP ANNPublication . Borghi, Pedro Henrique; Zakordonets, Oleksandr; Teixeira, João PauloWith the accelerated spread of COVID-19 worldwide and its potentially fatal effects on human health, the development of a tool that effectively describes and predicts the number of infected cases and deaths over time becomes relevant. This makes it possible for administrative sectors and the population itself to become aware and act more precisely. In this work, a machine learning model based on the multilayer Perceptron artificial neural network structure was used, which effectively predicts the behavior of the series mentioned in up to six days. The model, which is trained with data from 30 countries together in a 20-day context, is assessed using global and local MSE and MAE measures. For the construction of training and test sets, four time series (number of: accumulated infected cases, new cases, accumulated deaths and new deaths) from each country are used, which are started on the day of the first confirmed infection case. In order to soften the sudden transitions between samples, a moving average filter with a window size 3 and a normalization by maximum value were used. It is intended to make the model's predictions available online, collaborating with the fight against the pandemic.
- Optimization of glottal onset peak detection algorithm for accurate Jitter measurementPublication . Fernandes, Joana Filipa Teixeira; Borghi, Pedro Henrique; Freitas, Diamantino Silva; Teixeira, João PauloJitter is an acoustic parameter used as input for intelligent systems for the diagnosis of speech related pathologies. This work has the objective to improve an algorithm that allows to extract vocal parameters, and thus improve the accuracy measurement of absolute jitter parameter. Some signals were analyzed, where signal to signal was compared in order to try to understand why the values are different in some signal between the original algorithm and the reference software. In this way, some problems were found that allowed to adjust the algorithm, and improve the measurement accuracy for those signals. Subsequently, a comparative analysis was performed between the values of the original algorithm, the adjusted algorithm and the Praat software (assumed as reference). By comparing the results, it was concluded that the adjusted algorithm allows the extraction of the absolute jitter with values closer to the reference values for several speech signals. For the analysis, sustained vowels of control and pathological subjects were used.