Novel Low Coherence Interferometric Tools for Early Detection of Cancer

Abstract

Cancer encompasses many diseases affecting all parts of the body, is characterized by rapid, abnormal cell growth, and is one of the most common causes of death worldwide. Certain cancers, including cervical and colorectal cancer, exhibit very slow growth and histological evidence on or close to the tissue surface of precancerous development. With appropriate methods, the cancer can be detected before it has the chance to actually become malignant, and safely removed, preventing additional risk from more advanced disease. Developments in optical diagnostic techniques have shown promise in early detection of colorectal and cervical cancer, potentially able to improve upon existing preventative screening techniques in terms of accuracy and detection time.This dissertation is focused on optical methods based on low-coherence interferometry towards early detection of cervical and colorectal cancer. Angle-resolved low coherence interferometry (a/LCI) was used to build a system with scanning capabilities that was used for in vivo assessment of cervical dysplasia in a clinical study involving 20 patients. The results and prospective comparison to a previous a/LCI study had high accuracy (85%) as well as consistent numerical bright lines for normal vs. dysplastic tissue, showing that it is a promising technique towards diagnosing cervical dysplasia.Spectroscopic optical coherence tomography (SOCT) is another technique based on low-coherence interferometry that was investigated towards early detection of colorectal cancer (CRC). The large amount of genetic contributions to adenomatous polyps result in a large diversity of polyp morphologies, each with their own unique malignant potential. The depth-resolution capabilities as well as rich spectral information available from SOCT allows for computation of optical biomarkers such as scattering attenuation coefficient (????) or scattering power (SP) to be determined by imaging tissue, leading to quantitative parameters that can define different morphologies of adenoma. Mouse models representing different severities of CRC were imaged with SOCT and separated with high accuracy. Deep learning methods were applied to both mouse and subsequent human data with high accuracy, with benign human tissues separable from tubular and tubulovillous adenomas with 93.1% accuracy.