Tissue intrinsic emission fluorescence provides useful diagnostic information for various diseases. Because of its unique feature of spectral profiles depending on tissue types, spectroscopic imaging is a promising tool for accurate evaluation of endogenous fluorophores. However, due to difficulties in discriminating those sources, quantitative analysis remains challenging. In this study, we quantitatively investigated spectral-spatial features of multi-photon excitation fluorescence in normal and diseased livers. For morphometrics of multi-photon excitation spectra, we examined a marker-controlled segmentation approach and its application to liver fibrosis assessment by employing a mouse model of carbon tetrachloride (CCl4)-induced liver fibrosis. We formulated a procedure of internal marker selection where markers were chosen to reflect typical biochemical species in the liver, followed by image segmentation and local morphological feature extraction. Image segmentation enabled us to apply mathematical morphology analysis, and the local feature was applied to the automated classification test based on a machine learning framework, both demonstrating highly accurate classifications. Through the analyses, we showed that spectral imaging of native fluorescence from liver tissues have the capability of differentiating not only between normal and diseased, but also between progressive disease states. The proposed approach provides the basics of spectroscopy-based digital histopathology of chronic liver diseases, and can be applied to a range of diseases associated with autofluorescence alterations.
Keywords: autofluorescence; digital pathology; image segmentation; liver fibrosis; multi-photon microscopy; spectral imaging.