Statistical evaluation of fluorescence tumour spectral features through excitation-emission matrices for gastrointestinal cancer detection

The sensitivity of the autofluorescence spectroscopy for morphological and biochemical tissue alterations has been implemented as a “red flag” technique for detection of gastrointestinal tumours. Although this modality has proven beneficial, especially for less experienced physicians, the full capacity of the autofluorescence spectroscopy as diagnostic modality for cancer detection is yet to be established. This requires better understanding of the diagnostic value of the observed spectral properties differing cancerous and healthy gastrointestinal tissues, also defining highly specific optical markers based on these spectral peculiaritiesThis work presents an investigation of fluorescence characteristics of ex vivo gastrointestinal tissue samples of cancerous and healthy tissue, through excitation-emission matrices (EEMs). This method allows detection of fluorescence pattern in a broad spectrum range along with evaluating the correlation of diagnostically valuable fluorophores’ fluorescence maxima and the respective excitation wavelengths. The main tissue fluorophores, contributing for the observed EEMs, are tyrosine, tryptophan, NADH, FAD, collagen and elastin.Since the EEMs produce a lot of data, principal component analysis was used to reduce them to the most meaningful spectral features in respect to excitation and emission wavelengths. Obtained parameters were tested and the most distinguishable ones were implemented in support vector machine classifier to evaluate their performance for differentiation between cancerous and healthy gastrointestinal tissue. The aim of this study is to find robust optical parameters that could be a basics of highly feasible diagnostic algorithm for prospective clinical application.