Área do cabeçalho
gov.br
Portal da UFC Acesso a informação da UFC Ouvidoria Conteúdo disponível em:PortuguêsEnglishEspañol
Brasão da Universidade Federal do Ceará

Universidade Federal do Ceará
Mestrado e Doutorado em Ciências da Computação

Área do conteúdo

Defesa de Proposta de Mestrado: Antonio Guilherme Cunha Santos

Data da publicação: 6 de novembro de 2024 Categoria: Notícias, Proposta de Dissertação

Título: Enhanced Prognostic Modeling for Post-Cardiac Arrest Patients Using Kolmogorov-Arnold Networks as a Complementary Approach to EEG Signal Classification

Data: 11/11/2024
Horário: 14h00
Local: Sala de Seminários – Bloco 952

 

Resumo:

According to DATASUS, approximately 13,000 cardiac arrests occurred in Brazil between 2020 and 2023. Electroencephalogram (EEG) has proven valuable in the post-cardiac arrest prognosis, helping to assess neurological sequelae through the Cerebral Performance Category (CPC) scale, which ranges from full recovery (CPC = 1) to severe impairment or death (CPC = 5). This dissertation proposes an innovative approach that combines ConvNeXt (a modernized convolutional neural network (ConvNet) architecture) with Kolmogorov-Arnold Networks (KAN) to enhance the accuracy of prognostic predictions in post-cardiac arrest patients. In this approach, ConvNeXt will serve as the primary model for feature extraction and initial outcome predictions across sequential 12-hour EEG segments, covering the period from 1 to 72 hours after ROSC (return of spontaneous circulation). The KAN network will then act as an additional layer, utilizing the outputs of ConvNeXt across these time windows to capture temporal relationships and refine the prognostic predictions. Furthermore, burst suppression patterns will be included as an experimental feature to evaluate the hypothesis that combining these attributes with ConvNeXt’s architecture can improve prognostic capabilities. By leveraging the combined strengths of ConvNeXt for accurate and scalable feature extraction and KAN for temporal learning and interpretability, this research aims to develop a robust prognostic model. The ultimate goal is to provide clinicians with a more reliable and interpretable tool for making informed decisions about patient outcomes following cardiac arrest, thereby improving post-resuscitation care and quality of life for survivors.

Banca examinadora:

  • Prof. Dr. João Paulo do Vale Madeiro (MDCC/UFC – Orientador)
  • Prof. Dr. Luís Otávio Rigo Júnior (UNILAB – Coorientador)
  • Prof. Dr. César Lincoln Cavalcante Mattos (MDCC/UFC)
  • Prof. Dr. Manoel Alves Sobreira Neto (UFC)

 

Logotipo da Superintendência de Tecnologia da Informação
Acessar Ir para o topo