Normalized Raman Imaging (NoRI) enables high-resolution, label-free quantification of protein and lipid concentrations in biological tissues. Because NoRI provides rich molecular information, the analysis of its large, multi-channel datasets turns into a significant computational bottleneck. In this work, we introduce a novel, modular computational pipeline for automated segmentation and quantification of kidney tissue structures imaged with NoRI. The pipeline integrates classical image processing with state-of-the-art machine learning tools, including the Segment Anything Model (SAM) and ilastik, to segment key anatomical and biochemical features—such as tubules, nuclei, brush borders, and lumens. A custom contrast-enhancement strategy was developed to create a third SAM input channel from NoRI data, leading to a substantial improvement in segmentation performance (F1 score: 0.9226). Our framework enables accurate cytoplasm-resolved quantification of protein and lipid concentrations and reveals distinct biochemical signatures across renal tubule subtypes and experimental conditions. This method offers a robust, scalable foundation for quantitative tissue analysis and enhances the utility of NoRI imaging for biomedical research.
