CNN-based demosaicing for labeled fluorescence cancer intraoperative imaging with visible-NIR sensors

Abstract

Single-chip imaging devices with vertically stacked photodiodes and pixelated spectral filters enhance multi-dye imaging techniques for cancer surgeries. However, this advancement sacrifices spatial resolution. To address this issue, we have created a deep convolutional neural network designed to demosaic color and NIR channels, and its effectiveness has been confirmed through testing on both preclinical and clinical datasets. Keywords: CNN, demosaicing, near infrared imaging, image guided surgery, cancer surgery