Publicação

Classification of aortic stenosis based on AI in MRI scans

Ver documento

Detalhes bibliográficos
Resumo:Aortic stenosis (AS) stands as a significant cardiovascular ailment necessitating accurate diagnosis for effective patient management. This study introduces an innovative AI-based approach for AS detection in MRI scans. Our research aims to find a robust CNN model combined with computer vision techniques for the classification of AS in MRI, further refined through fine tuning. We evaluated five CNN models combined with computer vision techniques, where VGG16 model got the best results in our research work, with 95% in recall and 95% in F1-score. In this test four Data Augmentation techniques were implemented including Translation, Rotation, Flip and Brightness, enhancing the model’s robustness and generalization, encompassing real-world image variations encountered in clinical settings. This validation reaffirms the model's clinical applicability, promising streamlined diagnostics while allowing medical professionals to focus on intricate decision-making and personalized care. In conclusion, our study underscores the potential of AI-driven AS detection in MRI. The merger of transfer learning and data augmentation yields high accuracy rates, validated in real clinical cases, signifying a significant advancement in precise cardiovascular diagnosis.
Autores principais:Águas, Pedro Miguel Ferreira Viegas
Assunto:MRI imaging techniques Aortic disease classification Inteligência artificial -- Artificial intelligence Deep learning Técnicas de imagem por RM Classificação de doenças da aorta
Ano:2023
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:ISCTE
Idioma:inglês
Origem:Repositório ISCTE
Descrição
Resumo:Aortic stenosis (AS) stands as a significant cardiovascular ailment necessitating accurate diagnosis for effective patient management. This study introduces an innovative AI-based approach for AS detection in MRI scans. Our research aims to find a robust CNN model combined with computer vision techniques for the classification of AS in MRI, further refined through fine tuning. We evaluated five CNN models combined with computer vision techniques, where VGG16 model got the best results in our research work, with 95% in recall and 95% in F1-score. In this test four Data Augmentation techniques were implemented including Translation, Rotation, Flip and Brightness, enhancing the model’s robustness and generalization, encompassing real-world image variations encountered in clinical settings. This validation reaffirms the model's clinical applicability, promising streamlined diagnostics while allowing medical professionals to focus on intricate decision-making and personalized care. In conclusion, our study underscores the potential of AI-driven AS detection in MRI. The merger of transfer learning and data augmentation yields high accuracy rates, validated in real clinical cases, signifying a significant advancement in precise cardiovascular diagnosis.