Building a Single Cell Database from Whole Slide Images

Abstract

[EN] Since the development of the Papanicolaou Test, cytopathology has been increasing in efficiency by demonstrating its capability to detect cancer precursors and malignant changes, in every stage of a disease, analyzing the cells in the patient sample. The cytologic diagnostic is able to cell labeling using the standardized PAP stains, even before it can be detected under a microscope by a cytopathologist. However, the diagnosis could be more accurate by designing automatic cytologic analyzers since more cells will be checked. In this bachelor’s thesis, a protocol was designed using real patient samples that can help detect cancer precursors through the alignment of Whole Slide Images, using the upper left corner and the cells. The protocol aims to build the single-cell database obtained from the Brightfield in WSI and the Digital Holographic Microscopy images with the annotations of the five biggest cells in each slide. Due to the wide volume of cells stored, the protocol could be used in other WSI due to the PAP classification according to displasia applied. The data samples are provided by a clinical study which was performed in a cooperation of the Institute of Pathology and different TUM’s Departments and Chairs. [1]. The single-cell database gained from this measurement scheme and protocol can also be used to develop and design algorithms and Deep Learning methods to classify cells and detect pre-cancerous cells in the future in a simpler way. Since the two microscopes implemented, the BF and the DHM, work differently, they will not be identical and a displacement in the alignment of cells is expected. Therefore, the normalized error from these cells is estimated. However, even if it should not be a big error, the slide is very sensitive since it works with micrometers, µm .

[ES] Para proporcionar la posibilidad de crear un conjunto de datos de verdad para la citología, se llevó a cabo un estudio de referencia menor con 88 muestras de pacientes en colaboración con la Cátedra Heinz-Nixdorf de Electrónica Biomédica, el Instituto de Patología, Patología Computacional y el Departamento de Ginecología y Obstetricia de la TUM en el Hospital Universitario Rechts der Isar. Las diapositivas de los frotis cervicales teñidos de PAP se escanearon digitalmente en su conjunto utilizando un microscopio de campo brillante y el microscopio holográfico digital.

El objetivo de este trabajo es crear una base de datos que contenga celdas individuales con imágenes con un campo brillante y un microscopio holográfico digital a partir de imágenes de diapositivas completas y establecer un protocolo para construir una base de datos de este tipo a partir de la nueva WSI en el futuro. Los dos esquemas de imagen proporcionan cuatro versiones de una sola celda: la celda en campo brillante, el holograma, la amplitud y la imagen de fase cuantitativa.