Publicação
Detection and segmentation of macrophages in Quantitative Phase Images by Deep Learning using a Mask Region-based Convolutional Neural
| Resumo: | Quantitative Phase Imaging (QPI) has been demonstrated to be a versatile tool for minimally invasive label-free imaging of biological specimens and time-resolved cellular analysis. RAW 264.7 mouse macrophages were imaged by Digital Holographic Microscopy (DHM), an interferometry-based variant of QPI, in toxicological studies and cellular growth experiments. Robust detection and segmentation of cells in QPI images by Deep Learning facilitates automated data evaluation of images in high throughput microscopy. Detection, segmentation and the subsequent analysis of single cellular specimens in QPI images yields essential toxicity related physical parameters like the dry mass on the single-cell level. Deep Learning models, such as the Mask Region-based Convolutional Neural Network (Mask R-CNN), were proven to achieve robust results for object detection in fluorescence microscopy images. Thus, a Mask R-CNN was applied with the aim to obtain deeper cellular knowledge from DHM QPI images. This work shows that the combination of label-free DHM and a state-of-the-art Deep Learning model achieves reliable machine-generated data on the single-cell level and prospects to enhance the information as well as the quality of physical data that can be extracted from QPI images of biomedical experiments and label-free high throughput microscopy. |
|---|---|
| Autores principais: | Kutscher, Tobias |
| Assunto: | Deep Learning Mask R-CNN Digital Holographic Microscopy Quantitative Phase Imaging Biomedical Imaging Cell Detection and Segmentation |
| Ano: | 2021 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Quantitative Phase Imaging (QPI) has been demonstrated to be a versatile tool for minimally invasive label-free imaging of biological specimens and time-resolved cellular analysis. RAW 264.7 mouse macrophages were imaged by Digital Holographic Microscopy (DHM), an interferometry-based variant of QPI, in toxicological studies and cellular growth experiments. Robust detection and segmentation of cells in QPI images by Deep Learning facilitates automated data evaluation of images in high throughput microscopy. Detection, segmentation and the subsequent analysis of single cellular specimens in QPI images yields essential toxicity related physical parameters like the dry mass on the single-cell level. Deep Learning models, such as the Mask Region-based Convolutional Neural Network (Mask R-CNN), were proven to achieve robust results for object detection in fluorescence microscopy images. Thus, a Mask R-CNN was applied with the aim to obtain deeper cellular knowledge from DHM QPI images. This work shows that the combination of label-free DHM and a state-of-the-art Deep Learning model achieves reliable machine-generated data on the single-cell level and prospects to enhance the information as well as the quality of physical data that can be extracted from QPI images of biomedical experiments and label-free high throughput microscopy. |
|---|