Publicação
Automatic quantification of microglial cells from brain images
| Resumo: | Microglia are a type of glial cell residing in the central nervous system and represent about 10 to 15% of the brain cell population. These cells don’t produce electrical impulses and are responsible for fundamental physiological and pathological processes, as they represent the first line of immune defence within the central nervous system. Thus, the quantification of these cells is essential in a clinical context, as it allows better monitoring and planning of treatments for different pathologies. Conventional cell counting involves a specific set of tools and devices developed for this purpose. This process is time-consuming and imprecise due to being heavily dependent on the operator. Currently, most processes are performed manually. However, other approaches have been studied and developed to improve the counting process, making it less time-consuming, more efficient and reduce the error associated with factors external to the counting. That said, the objective of this dissertation is to study the best approach to automate the quantification of microglial cells, ranging from classical to deep learning methodologies. Combined with the appropriate image processing and analysis techniques, the classical approach proves to be an adequate solution. However, in recent years, approaches based on deep learning have shown promising performance in various image analysis tasks, such as classification, detection and segmentation. The approaches developed to automate the quantification process were tested on a set of images built in partnership with researchers from the School of Medicine of the University of Minho. As for the classical methodology approach, a protocol was developed within ImageJ, which was combined with image processing techniques that allowed the automation of the counting process. Based on Convolutional Neural Networks, the classification problem referring to a deep learning methodology obtained an accuracy of 0.9021 and managed to classify the 661 images in 5 minutes and 44 seconds. The two approaches, considered optimal within each methodology, are competitive with the state-of-the-art methods, as they allowed for the automation of the quantification process, and showed a significant improvement in reproducibility, efficiency and reduced error associated with human factors. |
|---|---|
| Autores principais: | Lopes, Diogo Alexandre Rodrigues |
| Assunto: | Automatic quantification of cells Central nervous system Deep learning Image processing Image segmentation Microglial cells Células microgliais Processamento de imagem Quantificação automática de células Segmentação de imagem Sistema nervoso central Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Microglia are a type of glial cell residing in the central nervous system and represent about 10 to 15% of the brain cell population. These cells don’t produce electrical impulses and are responsible for fundamental physiological and pathological processes, as they represent the first line of immune defence within the central nervous system. Thus, the quantification of these cells is essential in a clinical context, as it allows better monitoring and planning of treatments for different pathologies. Conventional cell counting involves a specific set of tools and devices developed for this purpose. This process is time-consuming and imprecise due to being heavily dependent on the operator. Currently, most processes are performed manually. However, other approaches have been studied and developed to improve the counting process, making it less time-consuming, more efficient and reduce the error associated with factors external to the counting. That said, the objective of this dissertation is to study the best approach to automate the quantification of microglial cells, ranging from classical to deep learning methodologies. Combined with the appropriate image processing and analysis techniques, the classical approach proves to be an adequate solution. However, in recent years, approaches based on deep learning have shown promising performance in various image analysis tasks, such as classification, detection and segmentation. The approaches developed to automate the quantification process were tested on a set of images built in partnership with researchers from the School of Medicine of the University of Minho. As for the classical methodology approach, a protocol was developed within ImageJ, which was combined with image processing techniques that allowed the automation of the counting process. Based on Convolutional Neural Networks, the classification problem referring to a deep learning methodology obtained an accuracy of 0.9021 and managed to classify the 661 images in 5 minutes and 44 seconds. The two approaches, considered optimal within each methodology, are competitive with the state-of-the-art methods, as they allowed for the automation of the quantification process, and showed a significant improvement in reproducibility, efficiency and reduced error associated with human factors. |
|---|