Paralelna faktorska analiza fluorescentnih svojstava višekomponentnih sistema

Abstract

Kombinacija optičkih spektroskopskih tehnika i odgovarajućih metoda za analizu i modeliranje spektralnih podataka se pokazala kao veoma obećavajuća tehnika za karakterizaciju i analizu organskih i neorganskih kompleksnih sistema. Razumevanje fizičkih fenomena koji se javljaju u kompleksnim sistemima predstavlja bitan preduslov za primenu spektroskopskih tehnika u različitim oblastima nauke. U ovoj tezi su izmerene fluorescentne karakteristike četiri kompleksna sistema: tri organska (rastvor vode i aminokiselina, med i tkivo dojke) i jedan neorganski (kompleksni sistem fosfora na bazi retkih zemalja). Dobijeni fluorescentni podaci su modelirani primenom paralelne faktorske analize. Izgrađeni modeli su omogućili definisanje fluorescirajućih komponenti (fluorofora) prisutnih u sistemima, pri čemu je model kao rezultat dao čiste spektre svake pojedinačne fluorofore, kao i njihove relativne koncentracije u svakom od uzoraka sistema. U neorganskom sistemu i rastvoru aminokiselina broj fluorofora i njihove koncentracije su unapred bile poznate i na taj način je testirana uspešnost modela kao i njegova sposobnost predikcije prisustva i koncentracija fluorofora u nepoznatim uzorcima. Pokazano je da je dobijeni model veoma uspešan, sa odstupanjem od samo 0.036% od stvarne koncentracije kod neorganskog sistema. Za sistem meda, model je pokazao da je u sistemu prisutno 6 fluorofora i dobijeni su čisti spektri svake od njih kao i njihove relativne koncentracije u svakom pojedinačnom uzorku. Dalje analize spektara meda su ukazale da je na osnovu koncentracija fluorofora moguće odrediti botaničko poreklo meda. Spektri meda su zatim analizirani metodom glavnih komponenata kao i diskriminantnom analizom parcijalno najmanjih kvadrata. Rezultati su pokazali veliku uspešnost primene metode fluorescentne spektroskopije za klasifikaciju i autentifikaciju uzoraka meda. Proračunom modela tkiva dojke je ustanovljeno da postoje 4 dominantne fluorofore. Utvrđeno je da je na osnovu njihovih koncentracija moguće sa tačnošću od 100% odrediti vrstu promena na tkivu (benigno ili maligno). Rezultati su potom poređeni sa rezultatima analize potpornih vektora primenjenih na neobrađene spektre, na osnovu kojih je pokazano da se sa istim podacima ne može dobiti podjednako uspešan model...

The combination of optical spectroscopic techniques and appropriate methods for analysis and modeling of spectral data proved to be a very promising technique for the characterization and analysis of organic and inorganic complex systems. Understanding of physical phenomena that occur in complex systems is an essential prerequisite for the application of spectroscopic techniques in various fields of science. Fluorescence characteristics of four complex systems: three organic (solution of water and amino acids, honey and breast tissue) and one inorganic (a complex system of phosphorus-based rare earth) were measured in this thesis. The obtained fluorescence data were modeled using parallel factor analysis. The constructed models have enabled definition of the fluorescing components (fluorophores) present in the system, wherein the model as a result gave pure spectra of the individual fluorophore as well as their relative concentrations in each sample of the system. In the inorganic system and solution of amino acids, the number of fluorophores and their concentrations were known in advance, and thus the performance of the model was tested as well as its ability to predict the presence and concentrations of fluorophores in the unknown samples. It was shown that the resulting model is very successful, with a deviation of only 0.036% of the actual concentration in inorganic systems. For the system of honey, the model showed that 6 fluorophores are present in the system and their pure spectra were obtained, as well as their relative concentrations in each sample. Further analysis of the spectra of honey have demonstrated that based on the concentrations of the fluorophores the botanical origin of honey can be determined. Spectra of honey were then analyzed using principal component analysis and discriminant partial least squares. Results have shown great success in application of fluorescence spectroscopy methods for classification and authentication of honey samples. Based on calculated model of the breast tissue it was determined that there are four dominant fluorophores. It was estabilshed that based on their concentrations it possible to determine the type of change in the tissue (benign or malignant) with an accuracy of 100%. The results were then compared with the results of support vector machine analysis applied to the raw spectra, from which it was shown that using the same data successful model can not be obtained (classification error 35.71%). However, it has been shown that with...