Evaluating collagen morphology and pathological lipid deposition using multiphoton image statistics

In this study we present a novel image analysis methodology to quantify and to classify morphological details in tissue collagen fibril organization and lipid deposition. Co-localized collagen (second harmonic, SHG) and lipid (coherent Raman, CARS) images of atherosclerotic artery walls were acquired by a supercontinuum-powered multi-modal nonlinear microscope. Textural features based on the first-order statistics (FOS) and gray level co-occurrence matrix (GLCM) parameters were extracted from the SHG and CARS images. Multi-group classifications based on support vector machine of SHG and CARS images were subsequently performed to investigate the potential of texture analysis in providing quantitative descriptors of structural and compositional changes during disease progression. Using a rabbit model, different collagen remodeling and lipid accumulation patterns in disease tissues can be successfully tracked using these image statistics, thus providing a robust foundation for classification. When the variation of the CARS image features were tracked against the age of the rabbit, it was noticed that older animals (advanced plaques) present a more complex necrotic core containing high-lipid extracellular structures with various shapes and distribution. With combined FOS and GLCM texture statistics, we achieved reliable classification of SHG and CARS images acquired from atherosclerotic arteries with >90% accuracy, sensitivity and specificity. The proposed image analysis methodology can also be applied in a wide range of applications to evaluate conditions involving collagen re-modeling and prominent lipid accumulation.