Publicação
Neural Radiance Fields (NeRFs) in Orthopedics
| Resumo: | Neural Radiance Field (NeRF) is a recent and promising deep-learning framework for view synthesis and three-dimensional (3D) reconstruction that uses deep neural networks for encoding the three-dimensional environment. Furthermore, Neural Radiance Fields have demonstrated dramatic improvements when compared to computer vision and computer graphics techniques using geometry or traditional machine learning tools. Despite the remarkable efficiency and potential of Neural Radiance Fields, there remains a notable scarcity of literature and research that explores the application of these techniques to the field of medical applications. In this thesis, we are interested in using Neural Radiance Fields for orthopedic procedures, specially focusing in arthroscopic video. This focus is driven by the unique challenges posed by the arthroscopic environment that is usually composed of low-textured structures (e.g., bones and soft tissues), large viewpoint changes, occlusions due to surgical instruments in front of the anatomies, specularities, floating debris and blood. All these concerns complicate view synthesis and three-dimensional reconstruction in arthroscopic environments, and there is presently no algorithm that performs these tasks accurately. The objective of this thesis is to explore the possibility of overcoming the above described difficulties using Neural Radiance Fields, and to perform view synthesis and three-dimensional reconstruction in arthroscopic environments in an accurate and robust manner. |
|---|---|
| Autores principais: | Moreira, Sofia Rebelo de Almeida |
| Assunto: | Neural Radiance Fields Arthroscopy Campo de Radiação Neuronal Artroscopia |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Coimbra |
| Idioma: | inglês |
| Origem: | Estudo Geral - Universidade de Coimbra |
| Resumo: | Neural Radiance Field (NeRF) is a recent and promising deep-learning framework for view synthesis and three-dimensional (3D) reconstruction that uses deep neural networks for encoding the three-dimensional environment. Furthermore, Neural Radiance Fields have demonstrated dramatic improvements when compared to computer vision and computer graphics techniques using geometry or traditional machine learning tools. Despite the remarkable efficiency and potential of Neural Radiance Fields, there remains a notable scarcity of literature and research that explores the application of these techniques to the field of medical applications. In this thesis, we are interested in using Neural Radiance Fields for orthopedic procedures, specially focusing in arthroscopic video. This focus is driven by the unique challenges posed by the arthroscopic environment that is usually composed of low-textured structures (e.g., bones and soft tissues), large viewpoint changes, occlusions due to surgical instruments in front of the anatomies, specularities, floating debris and blood. All these concerns complicate view synthesis and three-dimensional reconstruction in arthroscopic environments, and there is presently no algorithm that performs these tasks accurately. The objective of this thesis is to explore the possibility of overcoming the above described difficulties using Neural Radiance Fields, and to perform view synthesis and three-dimensional reconstruction in arthroscopic environments in an accurate and robust manner. |
|---|