Desenvolvimento de um framework Python orientado à contribuição para análise e processamento de biossinais: exemplo de aplicação para eletroencefalogramas

• Ano de defesa: 2024
• Autor(a) principal: Hauer, Arthur
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Engenharia Elétrica e Informática Industrial; UTFPR
• Palavras-chave em Português: Processamento de sinais; Engenharia biomédica; Python (Linguagem de programação de computador); Eletroencefalografia; Software - Desenvolvimento; Signal processing; Biomedical engineering; Python (Computer program language); Electroencephalography; Computer software - Development; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/36087

In Biomedical Engineering applications, biosignal processing is often tied to tools already established in the market (typically closed to external contributions), open-source tools with complex architectures, or the costly effort of programming the processing pipeline for the problem at hand. Furthermore, programming the processing independently increases the likelihood of errors, potentially compromising the experiment and its results. In the case of open-source tools with complex architectures, the issue remains the same as previously mentioned, as complexity hinders the ability to contribute. A noticeable trend in recent years in research involving, particularly, motor imagery using electroencephalography (EEG) signals is the use of Python and MATLAB languages. Python is a popular choice as it is an interpreted, high-level language, easy to use, and offers an extensive range of libraries for signal processing and machine learning. Thus, this work proposes a new Python-based framework for conducting experiments involving biosignals. The framework was developed using established software engineering practices, following the workflow of requirements collection, architecture, implementation, and testing. In each stage of this workflow, methods were applied to facilitate contributions to the project, enabling its expansion and widespread use. To make it accessible to researchers with minimal programming knowledge, the framework employs a node-oriented architecture, where each node represents a signal transformation, such as filtering, feature extraction, or classification. The user interface for configuration is a single, intuitively readable JSON file, making the experimental setup inherently replicable and contributing to research reproducibility. Additionally, the framework was validated through the experimental replication of a published work in the context of EEG classification, achieving a classification performance, measured by the area under the receiver operating characteristic curve, with a maximum of 0.847 and an average of 0.698, similar to the replicated work, with p between 0.008 and 0.945 obtained through the Wilcoxon signed-rank test grouped by individuals in the dataset, and p value of 0.339 without grouping, indicating similarity in results. Furthermore, during the execution of the replicated experiment, the framework consumed an average of 3228 MB of RAM, 104.48% CPU usage, and a runtime of 64.55 seconds on a AMD Ryzen 5 3600 processor.