Detection and classification of gastrointestinal cancer and other pathologies through quantitative analysis of optical coherence tomography data and goniophotometry

Abstract

The changes in light interaction between healthy and diseased tissues have been investigated as a potential diagnostic application. Here we attempt to differentiate between healthy and pathological gastrointestinal tissues using quantitative analysis of optical coherence tomography (OCT) data and goniophotometry. A goniophotometer was constructed and calibrated using titanium oxide and microsphere phantoms. Measurements were carried out on human gastrointestinal tissue sections collected using the methodology described below. The anisotropy factor g was extracted from the scattering curves by fitting the Henyey-Greenstein function. Measurements on human samples were in the forward scattering range with g 0.6-0.7, in agreement with the literature. Optical coherence tomography imaging was carried out on gastrointestinal tissues collected from patients undergoing elective surgery or endoscopy at St. Mary’s Hospital, London. In total 146 patients were included. Data was processed using gradient analysis of signal attenuation and morphological analysis with kNN classification. Results were correlated with histological diagnoses. Gradient analysis results were statistically significant across most categories, showing particularly good differences in the gradient distributions between healthy and diseased oesophageal tissues. Morphological analysis and kNN classification produced sensitivity and specificity values for healthy oesophagus and cancer in surgical specimens reaching 100% / 97.87% and 99.99% / 99.91% respectively and high accuracy in detecting Barrett's oesophagus in endoscopic specimens, with sensitivity and specificity values of 99.80% and 99.02%. Results in rectal tissue where also noteworthy, with detection of dysplasia reaching a sensitivity and specificity of 99.55% / 96.01%. Despite limitations in our work, we have shown that the detection of gastrointestinal pathologies using quantitative analysis of OCT data is a promising technique with good ex vivo results. Transferring the methodology to the in vivo domain holds a lot of potential as a future quick and reliable diagnostic technique.