Detection of cervical cancer from evoked tissue fluorescence images using 2- and 3-way methods

摘要

In-vivo examinations of cervical tissue were performed using Evoked Tissue Fluorescence (ETF) as part of aclinical trial. These examinations were performed in conjunction with conventional colposcopy, along with ofbiopsies of suspect cancerous or pre-cancerous tissue. The ETF data consisted of 22 images of the cervixtaken at combinations of 3 excitation wavelengths and 9 emission wavelengths. The ultimate goal was to usethese images to accurately classify tissue between normal states (normal squamous and normal columnar) andpre-cancerous and cancerous states (squamous metaplasia, and high and low Squamous Intraepithelial Lesion,HSIL and LSIL). Several chemometric methods were applied to the data for the purpose of preprocessing(e.g. image alignment), eliminating patient to patient variability (e.g. GLS and specialized centering),classification of tissues (hierarchical PLS-DA) and elucidating the underlying signatures of the tissues andbackground components (PARAFAC).