Leveraging Label-Free Multiphoton Microscopy for Diagnostic Applications: Probing Tissue Biology for Enhanced Disease Detection and Analysis

Abstract

Label-free optical imaging is a rapidly expanding field with growing diagnostic potential,fueled by advances in imaging technology and computational methods. This bioimaging method relies on light-tissue interactions to generate detectable contrast. These interactions can result in various physical phenomena, such as the emission of fluorescence from endogenous molecules and harmonic generation. The label-free contrast generated by these interactions is directly influenced by tissue state, where variations in the abundance and organization of contrast- producing molecules and structures lead to unique signatures. This is particularly relevant in cancer research and diagnostics, given the complexity of the disease. A model of this complexity can be found in gastroenteropancreatic neuroendocrine tumors (GEP-NETs), a highly heterogeneous cancer type that develops in the gastrointestinal tract. Optical imaging diagnostics have the potential to improve our ability to rapidly assess disease states and infer the biological characteristics of tumors, thus advancing personalized medicine (i.e., administering therapies tailored to specific tumor subtypes). This study explores the use of label-free optical imaging, particularly multiphoton microscopy (MPM), as a diagnostic tool using GEP-NETs as a disease model. In three key sections, we: 1) investigate the processing of MPM images to remove artifacts that hinder image analysis and feature extraction; 2) examine the use of MPM image features in fixed and fresh tissue to distinguish between normal and diseased tissue; and 3) combine advanced gene sequencing with MPM to validate existing and explore new image features that differentiate normal from diseased tissue and tumor subtypes. The results of Section 1 provide an open-source, easily implementable method for reducing tile artifacts in scanned MPM images, with suggestions for further improving this methodology. Section 2 shows that MPM image features in both fixed tissue and fresh tissue are qualitatively different between normal and cancerous tissue in GEP-NET cases, with quantitative analysis pending for fresh tissue. These features can be utilized in simple classifier algorithms to accurately distinguish tissue types, including tumor regions, suggesting potential future applications in computer-aided diagnostics. Finally, the study demonstrates that combining spatial gene sequencing with MPM imaging reveals correlations between known MPM features and regional changes in gene expression pathways. MPM and spatial gene features were also used to train deep-learning classifiers, which were able to classify MPM images into genetic clusters with high accuracy, indicating promising applications in optical genetic phenotyping.