Algoritmos computacionais para análise quantitativa de biomarcadores em imagens de medicina nuclear

Abstract

The present Dissertation belongs to the field of Computational Vision, more concretely to the segmentation and analysis of objects represented in images. While Computational Vision seeks to make decisions about real objects based on images through the construction of artificial systems, image segmentation and analysis seeks to construct models capable of effectively characterizing objects and segmenting into new images. This Dissertation intends to evaluate the response to multifocal bone metastasis therapy and to evaluate cardiac involvement in patients with suspected amyloidosis. With this purpose, computational algorithms were implemented for segmentation and image analysis for application in structures such as the axial skeleton, some appendicular regions and the myocardium. The aim of assessing the response to multifocal therapy comes from the need to quantitatively evaluate the response to prescribed therapy. Usually patients with metastatic bone cancer perform periodic bone scans (every 3 months) to assess the progression or regression of the disease. However, this analysis is only qualitative. The same is true in studies for the detection of cardiac amyloidosis, the evaluation is merely qualitative, and there is also a need to evaluate cardiac involvement quantitatively. Consequently, a methodology was developed to segment the regions described through images obtained in Nuclear Medicine, specifically bone scans with radiopharmaceuticals derived from 99mTc, based on deformable statistical models, namely shape models. The developed models allowed to segment the regions of interest and to effect this segmentation in new images. In addition, the use of such techniques allowed the extraction of quantitative measurements in those regions that were used to determine the rates of uptake of radiopharmaceuticals in hypercaptive foci, particularly metastases in metastatic cases and the myocardium in studies for the detection of amyloidosis. The results obtained by applying the methodology suggest that the efficiency of the model depends heavily on the image and the information that is being extracted from the regions. The good results obtained by the application of the active shape models for segmentation of skeletal regions into new images demonstrate that this type of model can be used for this purpose. With regard to the indices obtained, from an overall point of view, these were able to translate the response to the therapy of bone metastasis and to the detection of cardiac amyloidosis. In short, Computational View and object modeling can be used to support this function as demonstrated in this Dissertation. Thus, the developed model allows evaluating the response to the therapy of bone metastasis and in the detection of cardiac amyloidosis.

Date of Award	8 Jan 2018
Original language	Portuguese
Awarding Institution	Universidade Católica Portuguesa
Supervisor	João Manuel Ribeiro da Silva Tavares (Supervisor) & Diogo Alexandre Borges de Faria (Co-Supervisor)