National Technical University of Athens (NTUA)
Ανάλυση εικόνων μαστογραφιών για την ανίχνευση των μικροαποτιτανώσεων και την κατηγοριοποίησή τους με νευρωνικά δίκτυα
Abstract
dc:descriptionEarly detection is the key to improve breast cancer prognosis and the only proven effective method for this is mammography. An early sign of 30-50% of breast cancer incidents is the appearance of clusters of fine, granular microcalcifications, while 60-80% of breast carcinomas reveal microcalcification clusters upon histological examination. The extended correlation between the appearance of microcalcification clusters and breast cancer indicates that, by exploiting computer aided diagnosis (CAD) systems for automated classification of micro calcification clusters, we can acquire a very useful and helpful tool for breast cancer control. Microcalcifications appear in a mammogram as fine, granular clusters, which are often difficult to identify in a raw mammogram. A variety of techniques have been proposed to enhance and automatically detect microcalcifications, but there are still no satisfactory, clinically acceptable results. The essence of this dissertation is to present Matching Pursuit (MP) as a powerful technique to code texture and patterns in mammograms, due to its property of analyzing and simplifying them similarly to what the human brain would do. MP is a greedy algorithm that decomposes any signal into a linear expansion of waveforms that are selected from a redundant dictionary. The choice of the dictionary is a crucial point for the performance of this algorithm for an image coding. Here we have introduced the use of Gabor atoms dictionary, which will help to code texture and patterns in images. This technique, when working with a low amount of data, has good performance and a fast decay of the coding error. The output of the MP algorithm was used as an input in a tree-layer feedforward perceptron for the classification of regions of interest (ROIs). The performance of this neural network has shown that MP algorithm is a very promising tool in mammograms' analysis as the outcome is much better than the classification performance of other conventional techniques, wavelets or statistical methods for example.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2007
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Sakka, Eleni
- Σακκά, Ελένη
Subjects
dc:subject × 12- Μικροαποτιτανώσεις
- Δίκτυα, Νευρωνικά
- Μαστογραφίες
- Matching pursuit, Μετασχηματισμός
- Matching pursuit, Transform
- Microcalcifications
- Networks, Neural
- Mammographies
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/16880