An etalon-based mid-infrared probe can be configured for spectroscopic tissue discrimination, such as between non-normal (e.g., cancerous) and normal (e.g., healthy) tissue. A broadband light source can be applied to the etalon to generate fringes at spectroscopic wavelengths of interest, which can be delivered to a tissue specimen via a fiber loop probe. A response signal can be spectral dispersed across a parallel array of detector pixels, such as using a diffraction grating, and signal processed for performing the tissue classification. A learning model can be trained, using full IR spectral data, for applying a reduced set of wavelengths for performing the spectroscopic tissue analysis and classification.