Back to search

National Technical University of Athens (NTUA)

Ανάπτυξη μεθόδων επεξεργασίας εικόνων μαστογραφίας βασισμένων σε τοπικά μορφολογικά χαρακτηριστικά για την υπολογιστική υποστήριξη της διάγνωσης

Abstract

dc:description

In the current thesis we propose a series of medical image processing methods, based on local morphological characteristics, which can be utilized as building components on a computer aided diagnosis system (CAD). The common characteristic of these methods is the use of a tensor like feature which we call radial length. We first evaluate the use of tensor lengths on contrast enhancement. An enhancement process is used by the majority of CAD systems as a pre-processing step since it can significantly improve the performance of the whole system. Given the nature of mammographic images we attempt to perform contrast enhancement while preserving the diagnostically critical information. Therefore, in order to evaluate the performance of the proposed method we adopt an evaluation process which evaluates the contrast improvement achieved specifically between a lesion region (e.g. mass) and its surrounding tissue. The proposed method is then compared with other widely used methods outperforming them in 65.8% - 85.8% of cases depending on the method under comparison. Next, we evaluate the use of radial lengths for the segmentation and removal of pectoral muscle from medio-lateral oblique mammographic images. The proposed procedure involves the use of radial lengths in order to define a number of candidate image points for the muscle-breast edge, followed by a line expansion procedure using the identified image points as starting points. The expansion process is based again on the values of tensor lengths on the different image locations. The evaluation of the method is performed through a specifically defined, by previous works, dataset where the muscle edge is pre-defined. The evaluation results prove the effectiveness of the proposed methodology achieving low error rates for 68.6% of cases as compared to 53.5% - 59.5% achieved by the methods under comparison. We further evaluate the use of radial lengths for the identification of architectural distortions. Architectural distortions are the third most common form of mammographic lesions, behind masses and microcalcifications, but are the most difficult to identify. The proposed method tries to combine radial lengths with local binary patterns (LBP) and their modifications. Due to the large number of combinations used, the number of the produced characteristics is very large making the need for transforming them and reducing their size, imperative. In order to achieve this, we propose the combined use of principal components analysis (PCA) and t-test. The reduced characteristics set is then classified using a support vector machine (SVM) classifier with the results being very encouraging especially when large datasets are used. More specifically, the classification is flawless when the whole number of architectural distortions in DDSM is used. Furthermore, we propose an educational game for trainee radiologists in order to assist them in their efforts to effectively interpret mammograms. The main architectural characteristics of the game are defined after a thorough bibliographic analysis of modern learning theories and the way they can be applied on games.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2020

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Chatzistergos, Sevastianos
  • Χατζηστέργος, Σεβαστιανός

Subjects

dc:subject × 16

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/48070

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Chatzistergos, Sevastianos; Χατζηστέργος, Σεβαστιανός. Ανάπτυξη μεθόδων επεξεργασίας εικόνων μαστογραφίας βασισμένων σε τοπικά μορφολογικά χαρακτηριστικά για την υπολογιστική υποστήριξη της διάγνωσης. National Technical University of Athens (NTUA), 2020. http://hdl.handle.net/10442/hedi/48070