Author: Bc. Ivan Vykopal
Heart transplantation is a procedure that may result in cardiac transplant rejection, necessitating measures to prevent rejection. To evaluate the patient's health status ad aid in early diagnosis of allograft rejection, doctors perform numerous invasive procedures, such as biopsies, followed by histopathological analysis of the samples. The aim of the doctors is to use the samples to gain as much information as possible to evaluate the patient's health status, which gives a scope for the use of methods based on deep learning to help obtain additional quantitative information for early diagnosis of allograft rejection.
In this thesis, we propose a method aimed at quantitative analysis of histological images to make the work of doctors more efficient in diagnosing the patient's health condition. The proposed method consists of segmenting higher morphological structures, especially inflammation, using deep learning methods and traditional methods based on computer vision. Based on the proposed solution, we have designed and implemented several models and approaches for the segmentation of higher morphological structures. We evaluated the proposed methods from qualitative and quantitative perspectives on data provided by doctors and also in collaboration with doctors from the Institute for Clinical and Experimental Medicine in Prague. We also integrated the proposed models into the QuPath tool using the MONAI Label extension for potential use in the practice.