Comparative analysis of pigmented skin lesions and normal skin using optical spectroscopic methods

Optiĉka spektroskopija i imidţing tehnike mogu biti korisne za ranu dijagnostiku kancera. U ovoj tezi, razlike u luminescentnim odgovorima pigmentnih lezija (melanoma i nevusa) i po svemu sudeći nepigmentne koţe ljudi su analizirane sinhronom luminescentnom spektroskopijom. Merenja su izvršena u opsegu eksitacione talasne duţine od 330 do 545 nm i sinhronom intervalu od 30 do 120 nm, da bi se dobili osnovni i trodimenzionalni sinhroni luminescentni spektri (3DTSLS) za svaki uzorak. Normalna koţa, nevus i melanom se razlikuju u naĉinu na koji fluoresciraju, i te razlike su više naglašene u sinhronim fluorescentnim spektrima nego u klasiĉnim emisionim i eksitacionim spektrima. Razlike u fluorescentnim karakteristikama pigmentnih lezija i zdrave koţe potiĉu od razliĉitih koncentracija endogenih fluorofora i hromofora. Analizom glavnih komponenata i linearnom diskriminantnom analizom sinhronih spektara izmerenih za razliĉite sinhrone intervale pokazano je da je najveća razlika izmeĊu grupa uzoraka dobijena za sinhroni interval 70 nm. Ovi spektri su zatim korišćeni za pravljenje klasifikacionih modela diskriminantnom analizom parcijalno najmanjih kvadrata. Testiranjem uspešnosti ovih modela krivom operativnih karakteristika pokazano je da su oni optimalni, sa maksimalnom vrednošću od jedan za površinu ispod krive operativnih karakteristika za detekciju melanoma. Za testiranje primenjivosti pomenutih tehnika korišćene su veštaĉke neuronske mreţe. U ovu svrhu, samoorganizujuće mape (SOM) i višeslojni perceptron (MLP) su obuĉene na osnovu luminescentnih podataka, a zatim je izvršeno grupisanje uzoraka tkiva. Pokazano je da su obe mreţe SOM i MLP dale dobre rezultate sa stopom uspešnosti klasifikacije uzoraka tkiva od 96% i 98%. Drugi deo eksperimentalnog rada u ovoj tezi se sastojao od merenja difuzno refleksionih spektara koţe volontera na dve regije tela, ĉelu i šaci, u cilju analize svojstava boje koţe na objektivan naĉin. Dobijeni spektri su analizirani korišćenjem analize glavnih komponenata (PCA) i L*a*b* sistema boja koji je preporuĉen od MeĊunarodne komisije za osvetljenje (CIE). Na osnovu rezultata analiza pokazano je da difuzno refleksiona spektroskopija uz korišćenje analize glavnih komponenata i L*a*b* sistema boja predstavlja korisnu i objektivnu metodologiju za analiziranje karakteristika koţe.Novel optical spectroscopy and imaging methods could be valuable in the early detection of cancer. Differences in the luminescence responses of pigmented skin lesions (melanomas and nevi) and apparently non-pigmented human skin were analyzed using synchronous luminescence spectroscopy in this thesis. Measurements were performed in the excitation range of 330-545 nm, with synchronous interval varyng from 30-120 nm to obtain ordinary and three-dimensional total synchronous luminescence spectra (3DTSLS) of each specimen. Normal skin, nevi and melanomas differ in the way they fluorescence, and these differences are more distinct in the syncronous fluorescence spectra than in the conventional emission and excitation spectra. The differences in the fluorescence characteristics of pigmented and normal skin samples were ascribed to differences in concentracions of endogenous fluorophores and chromophores. Principal component and linear discriminant analysis of the synchronous spectra measured at different synchronous intervals showed that the greatest variance among the sample groups was at the 70 nm interval spectra. These spectra were then used to create partial least squares discriminant analysis-based classification models. Evaluation of the quality of these models from the recevier operating characteristic curves showed they performed well, with a maximum value of one for the area under the curve for melanoma detection. To test applicability of the presented techniques artificial neural networks were used. For this purpose self organizing maps (SOM) and multi layer feed-forward neural network (MLP) were trained with luminescence data and then used for tissue specimen clustering. It was shown that both SOM and MLP gave promising results with 96 to 98% success tissue classification rate. Second part of experimental work in this thesis were measurements of diffuse reflectance spectra from skin sites located on a forehead and top of a hand of healthy volunteers in order to analyze skin color properties in an objective way. Obtained spectra were analyzed using principal component analysis (PCA) and standard Commission International d'Eclairage (CIE) L*a*b* color scheme. Based on results of analysis it was shown that diffuse reflectance spectroscopy combined with the PCA and the L*a*b* color system is a convenient and objective methodology for analyzing skin properties