Development of hyperspectral microscopy for improving the diagnostic accuracy in leukemia diagnosisDevelopment of hyperspectral microscopy for improving the diagnostic accuracy in leukemia diagnosisΑνάπτυξη υπερφασματικού μικροσκοπίου για την βελτίωση της διαγνωστικής ακρίβειας στην διάγνωση λευχαιμίας Διπλωματική Εργασία Diploma Work 2016-05-272016enSpectral Imaging is a combination of Spectroscopy and Imaging. It is based on the acquisition of a collection of narrow spectral band images (Spectral cube). In Hematology, the May Grunwald-Giemsa (MG-G), a solution mixture of Eosin Y, Azure B and Methylene Blue stains, is commonly used for leukemia diagnosis in blood and bone marrow samples. Diagnosis is based on manual observation of stained tiles. This procedure is time consuming and subjective, as it depends on the experience and judgement of the observer-hematologist. In this diploma thesis, we present a novel Spectral Imaging Microscopy modality, for mapping the concentrations of MG-G substances. Partial Least Squares (PLS) regression algorithm is employed as most efficient in chemo-metrics. Using the spectral cube absorbance data the concentration of each stain is estimated per pixel. Using Dimensionality Reduction Analysis the minimum essential number of spectral bands is estimated, ensuring minimum loss in accuracy and much faster data analysis. Finally, four scaled concentration pseudo-color maps are produced, one for each substance of MG-G and one depicting the proportions of their concentrations, which are used for the identification of the leukemic cells, in an objective and reproducible diagnosis procedure.Diploma Thesishttp://creativecommons.org/licenses/by/4.0/Πολυτεχνείο Κρήτης::Σχολή Ηλεκτρονικών Μηχανικών και Μηχανικών ΥπολογιστώνGotsoulias_Dimitrios_Dip_2016.pdfChania [Greece]Library of TUC2016-05-27application/pdf12 MBembargo Gotsoulias Dimitrios Γκοτσουλιας Δημητριος Balas Costas Μπαλας Κωστας Garofalakis Minos Γαροφαλακης Μινως Zervakis Michalis Ζερβακης Μιχαλης Πολυτεχνείο Κρήτης Technical University of Crete Hyperspectral imaging