Graduate and Postdoctoral Studies
Multi-scale optical imaging techniques for early cancer detection in the gastrointestinal tract
Wednesday, April 12, 2017
to 12:30 PM
1060 BioScience Research Collaborative
Gastrointestinal (GI) cancers impose an enormous burden on the society worldwide and a significant proportion of this burden can be prevented through early cancer detection and treatment. Current screening and surveillance protocols rely primarily on conventional white light endoscopy, the accuracy and efficacy of which need to be improved. The main objective of this research is to develop and optimize novel multi-scale optical imaging modalities to improve detection of GI cancers with enhanced imaging performance and increased clinical ease of use at a low cost.
A modular video endoscope (MVE) was developed to combine widefield with high-resolution imaging modalities. Trimodal imaging, including standard white light imaging (WLI), vital-dye fluorescence imaging (VFI) and high-resolution microendoscopy (HRME), was enabled in a single endoscopic insertion. A pilot in vivo clinical trial showed that glandular architectural dysregulation, as visualized in VFI and HRME, was associated with cancer progression in Barrett’s esophagus (BE). The MVE/HRME platform was further evaluated for gastric cancer detection. In both ex vivo and in vivo pilot studies, early cancers were found to be highlighted by alterations in glandular patterns and nuclear morphology in VFI and HRME. Preliminary data in the in vivo trial suggested that the platform may be useful to detect additional advanced lesions, while the specificity needs to be improved.
A low-cost confocal HRME was developed to improve the axial performance of HRME with optical sectioning. By synchronizing a digital light projector (DLP) with the rolling shutter of a CMOS sensor, line-scanning confocal imaging was enabled in a compact design. Initial ex vivo validation in imaging squamous and columnar epithelium of mouse specimens demonstrated that optical sectioning improved the visualization of nuclear morphometry, especially in crowded regions with degraded image quality.
Automated analysis of HRME images was also explored to facilitate its clinical applications. In 58 in vivo colorectal HRME images, a set of clinically relevant features were quantified. A 3-feature model was developed through linear discriminate analysis to achieve a sensitivity and specificity of 91% and 89%, and an AUC of 0.94 in classification of neoplastic from non-neoplastic polyps.
The unique contributions of this research are the development of multi-scale imaging modalities with enhanced imaging performance and improved clinical ease of use. Computer-aided interpretation of clinical data was also investigated. These results can potentially contribute to improved early GI cancer detection, especially in community and low-resource settings.