Back to search

National Technical University of Athens (NTUA)

Ανάπτυξη μεθοδολογιών για την υποβοηθούμενη διάγνωση συμπλεγμάτων μικροασβεστώσεων του μαστού

Abstract

dc:description

In the present PhD thesis, we develop methodologies which may be integrated in a computer aided diagnosis system (CADx) for clusters of microcalcifications detected in mammographic images. Microcalcifications are tiny deposits of calcium that can be located anywhere inside the breast and their presence is associated to the existence of breast cancer. The development of the system includes various independent steps such as image preprocessing, aiming to the enhancement of image quality and the detection of important mammographic structures, the segmentation of the microcalcifications from the original image, in order to isolate the findings from the background, the extraction of features describing the findings and finally the application of feature selection and classification methods for the discrimination between benign and malignant findings. Thus, we present the methods implemented for each different stage of the computer aided diagnosis system. We performed the analysis of a microcalcifications cluster by extracting features concerning the morphology and the brightness of the findings, their distribution inside the cluster, the morphology of the cluster and the texture of the surrounding tissue. We also evaluated and compared different feature selection methods and classification schemes, in order to find the most efficient combination which leads to optimization of the classification results. After developing all subunits of the system, we proceed on its evaluation in order to investigate its efficacy and its potential to be adopted in daily clinical practice. For the evaluation of the system we used images from real-life cases provided by public databases. A subset of 22 cases from the MIAS database and 1715 cases from the DDSM database was used, achieving consequently objective results, as it is the greatest subset of images used in studies concerning the automated diagnosis of clusters of microcalcifications. Through our measurements, we extract valuable conclusions concerning the factors that influence the analysis of mammograms, we investigate the performance using subsets of cases of different properties, we compare different feature extraction methodologies in order to locate the best subsets of features for the discrimination of clusters and finally we compare the classification results of the system with the corresponding results obtained by the radiologists that analyzed the same dataset. The results are encouraging, since there are subsets of cases where we achieve high classification results (Az=0.91), while we outperformed the results achieved by the radiologists. Additionally, the classification results of the present study are comparable or even superior to those reported in the literature. Based on the previous results and conclusions, we investigate two new stages: the first concerns the enhancement of the initial region of interest aiming at the improvement of its contrast, while the second stage concerns the introduction of location features for the diagnosis of the finding. As far as the first stage is concerned, we implemented and compared different image enhancement techniques on subsets of mammograms of varying breast density and subtlety. The results of the comparison reveal that specific properties of the mammogram determine the need of different algorithms in order to optimize the diagnostic process. Concerning the second stage, we develop methodologies for the identification of important breast landmarks (breast periphery, pectoral muscle and nipple) in order to extract features that describe the relative location of the cluster inside the breast. These features are used to develop a mammographic atlas which presents regions of the breast with high occurrence of clusters of microcalcifications. We also introduce the use of probabilistic maps which reveal areas of the breast with increased probability for malignancy. The specific maps are used for the development of a classification model able to provide a priori risk estimation for a cluster of microcalcifications, based absolutely on information extracted from the relative location of the cluster inside the breast. At the last chapter, we use the conclusions that we extracted previously and we combine the proposed methodologies, in order to form a scenario for the use of the CADx system in the daily clinical practice, by simulating the interaction between the radiologists and the CADx system. The proposed scenario includes the use of the CADx system where the suggestions of the radiologist determine the selection of the most appropriate methodologies to analyze the region of interest. The mammographic atlas and the probabilistic maps are used to extract a risk estimation related absolutely to the location of the cluster. Additionally, an extra risk estimation is calculated which is based on the description of the finding following terms described in the BIRADS standard, while the age of the patient is also used to add extra diagnostic information. The proposed scenario is evaluated using a subset of 481 cases and the Az value achieved is 0.861. If the diagnoses of both the radiologists and the CADx system are combined, we achieve high levels of sensitivity (97.2%) and specificity (41.1%), improving consequently the performance of the radiologist without using the CADx system (sensitivity 93.5%, specificity 16.6%).

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2014

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Andreadis, Ioannis
  • Ανδρεάδης, Ιωάννης

Subjects

dc:subject × 33

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/38487

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

Andreadis, Ioannis; Ανδρεάδης, Ιωάννης. Ανάπτυξη μεθοδολογιών για την υποβοηθούμενη διάγνωση συμπλεγμάτων μικροασβεστώσεων του μαστού. National Technical University of Athens (NTUA), 2014. http://hdl.handle.net/10442/hedi/38487