Συντάχθηκε 11-10-2013 11:55
από Balasi Panagiota
Email συντάκτη:
Ενημερώθηκε:
-
Ιδιότητα: -.
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Σχολή Ηλεκτρονικών Μηχανικών και Μηχανικών Υπολογιστών
Πρόγραμμα Προπτυχιακών Σπουδών
ΠΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ
Γιακουμάκη Θεόδωρου - Μάριου
με θέμα
Καθολική βελτιστοποίηση για την ανάλυση δυναμικών οπτικών δεδομένων για την in vivo διάγνωση της επιθυλιακής νεοπλασίας
Global optimization analysis of dynamic optical data for the in vivo diagnosis of epithelial neoplasia
Τρίτη 15 Οκτωβρίου 2013, 11πμ
Αίθουσα Συνεδριάσεων, Κτίριο Επιστημών, Πολυτεχνειούπολη
Εξεταστική Επιτροπή
Αναπληρωτής Καθηγητής Κωσταντίνος Μπάλας (επιβλέπων)
Καθηγητής Κωσταντίνος Καλαϊτζάκης
Επίκουρος Καθηγητής Γεώργιος Χαλκιαδάκης
Abstract
In vivo, dynamic, contrast-enhanced optical imaging is a novel, non-invasive, method for the identification of epithelial tissue abnormalities. It was first used in the case of the cervical neoplastic epithelium. A mathematical model simulating the pharmacokinetics of the biomarker has been developed. Eight structural and functional biological parameters, correlating well with neoplasia progress, have been identified and with global sensitivity analysis were reduced to a set of four most identifiable. In the current thesis, global optimization analysis has been performed for the purpose of investigating whether the model parameters can be estimated, within adequate accuracy, by fitting of outputs of the model to dynamic optical signals. The capability to predict uniquely the full and the reduced set of the proposed parameters has been assessed by a parallel working software, that was created in order to automatically compare nine, equally constrained, state-of-the-art global optimization algorithms. Statistical analysis concluded that the unidentified experimental optical data converge adequately. This occurs more predominantly when the reduced set of the four most identifiable parameters is used, in agreement with global sensitivity analysis results. In particular, the most optimally performing algorithm is currently the differential evolution algorithm, offering parameter estimation statistical accuracy up to 99% in under 2.5 minutes per trial and for the reduced set. On such grounds, it can be reasonably argued that the solution of parameter estimation problem is substantially unique. Finally, fitting of experimental data to the outputs of the model allow the prediction of the identifiable tissue parameters, thus enabling for the first time in silico biopsy of neoplasia growth.
