Multimodal esophageal cancer imaging: establishing data processing techniques and assessing diagnostic sensitivity.

Abstract

Multimodal optical imaging techniques have generated significant interest for applications such as cancer detection, as combining complementary modalities could broaden the ability to detect early disease changes, as well as address patient-to-patient variability. However, there are challenges in determining how different imaging modalities may or may not complement one another and how best to capitalize on these advantages through computational analysis. We investigate the application of multimodal imaging for the purpose of detecting esophageal cancer, the sixth most deadly cancer in the world. To achieve this, we acquired multimodal optical imaging data-specifically autofluorescence, hyperspectral, polarized light, and optical coherence tomography (OCT)-from fresh human tissue samples obtained during upper endoscopy. Our analysis then addressed three key questions: Which individual modality best differentiates between healthy and cancerous tissues? How can data from these modalities be integrated to maximize discrimination? What computational methods are suitable for analyzing the resulting high-dimensional multimodal datasets? Our findings indicate that polarized light imaging (PLI) exhibits the strongest discriminatory power under these imaging conditions, with potential benefits observed from combining PLI and OCT in a multimodal approach.