Multimodal and multi-spectral microscopy with virtual staining based digital pathology using deep neural networks for label-free detection of cancer

Abstract

We report the development of multi-modal and multispectral microscopy, which is equipped with a multi-color light source, such as UV-A, blue, cyan, and white light. The developed device captures high-resolution multispectral autofluorescence (AF), bright-field and quantitative phase images for enhanced diagnostic precision. From these images, autofluorescence signals of key metabolic biomarkers, such as, nicotinamide adenine dinucleotide (NADH), flavin adenine dinucleotide (FAD), and porphyrins are extracted. Quantitative analysis of these signals’ yields ‘Redox ratio’, a sensitive biomarker for distinguishing healthy tissues from malignant ones. Quantitative phase imaging was performed using transport of intensity equation. The rich information extracted from the developed microscope was used for virtual staining based digital pathology using deep neural networks for label-free detection of cancer. The present system and method require only one or steps of sample preparation without staining. The results are compared with gold standard histopathology.