Infrared spectroscopy of macroscopic samples can be calibrated against reference analysis, such as lipid profiles acquired by gas chromatography, and serve as a fast, low-cost, quantitative analytical method. Calibration of infrared microspectroscopic images against reference data is in general not feasible, and thus spatially resolved quantitative analysis from infrared spectral data has not been possible so far. In this work, we present a deep learning-based calibration transfer method to adapt regression models established for macroscopic infrared spectroscopic data to apply to microscopic pixel spectra of hyperspectral IR images. The calibration transfer is accomplished by transferring microspectroscopic infrared spectra to the domain of macroscopic spectra, which enables the use of models obtained for bulk measurements. This allows us to perform quantitative chemical analysis in the imaging domain based on infrared microspectroscopic measurements. We validate the suggested microcalibration approach on microspectroscopic data of oleaginous filamentous fungi, which is calibrated toward lipid profiles obtained by gas chromatography and measurements of glucosamine content to perform quantitative infrared microspectroscopy.
Magnussen et al. (2025) studied this question.