Συντάχθηκε 18-07-2014 15:40
από Esthir Gelasaki
Email συντάκτη: egelasaki<στο>tuc.gr
Ενημερώθηκε:
-
Ιδιότητα: υπάλληλος.
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Σχολή Ηλεκτρονικών Μηχανικών και Μηχανικών Υπολογιστών
Πρόγραμμα Προπτυχιακών Σπουδών
ΠΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ
ΕΛΕΝΗΣ ΧΙΟΥ
με θέμα
Data Classification and Mapping in Optical Dynamic Contrast-Enhanced Imaging of Cervical Neoplasia
Δευτέρα 21 Ιουλίου 2014, 1:30μμ
Αίθουσα 137.Π39, Κτίριο Επιστημών, Πολυτεχνειούπολη
Εξεταστική Επιτροπή
Αναπληρωτής Καθηγητής Κώστας Μπάλας (επιβλέπων)
Καθηγητής Μιχάλης Ζερβάκης
Αναπληρωτής Καθηγητής Μιχαήλ Λαγουδάκης
Abstract
The main purpose of this study is the evaluation of temporal feature extraction methods from normal and abnormal cervical epithelium images acquired by Dynamic Contrast Enhanced Optical Imaging (DCE-OI). Dynamic optical data were recorded in vivo during agent-tissue interaction inducing a transient tissue whitening phenomenon, known as acetowhitening. The degree and duration of the optical phenomenon is associated with the lesion grade. The available data, observed on 64 patients and confirmed by in total 371 biopsy samples, include all possible cases of neoplasia grades. In order to reduce the data dimensionality and extract valuable information, an extensive number of feature extraction methods, including Wavelet Transformation (WT), Principal Component Analysis (PCA), Kernel Principal Component Analysis (KPCA), Piecewise Aggregate Approximation (PAA), Adaptive Piecewise Aggregate Approximation (APAA) and Symbolic Aggregate Approximation (SAX), have been implemented and (cross) validated with a 1-NN classifier. The results indicate that using a subset of the entire feature set, WT, PCA and KPCA methods present similar or better performance compared to using the entire feature set. Also, selection of the best features extracted from PCA and WT demonstrates better performance for some classification cases. Regarding the execution times, PCA presents better performance than the others methods.