A deep learning approach to the measurement of the nucleus-to-cytoplasm ratio

<p>Cancer is commonly diagnosed by experts examining tissue samples using a microscope. Cellular features (e.g., nucleus) can be isolated and examined using dyes. Unusually large or small nuclei are a characteristic of many types of cancerous cells, leading to the development of the nucleus-to-cytoplasm ratio. Manually assessing the nucleus-to-cytoplasm ratio tends to be biased.</p> <p>The nucleus-to-cytoplasm ratio can be assessed using imaging flow cytometry images and fluorescent dyes. Imaging flow cytometry can be used to quickly image large numbers of cells, making it an ideal candidate for use with deep learning. Deep learning and imaging flow cytometry have been used for tasks such as classifying cells into mitotic phases.</p> <p>This thesis aims to use deep learning to assess the nucleus-to-cytoplasm ratio using imaging flow cytometry brightfield images. Examining whether the nucleus-to-cytoplasm ratio can be assessed without fluorescent dyes. Results suggest that deep learning can successfully assess the nucleus-to-cytoplasm ratio using brightfield images, with results comparable to imaging flow cytometry analysis, producing AUC values above 0.9.</p>