Abstract
Polarization-resolved second harmonic generation (pSHG) is a label-free method that has been used in a range of tissue types to describe collagen orientation. In this work, we develop pSHG analysis techniques for investigating cranial bone collagen assembly defects occurring in a mouse model of hypophosphatasia (HPP), a metabolic bone disease characterized by a lack of bone mineralization. After observing differences in bone collagen lamellar sheet structures using scanning electron microscopy, we found similar alterations with pSHG between the healthy and HPP mouse collagen lamellar sheet organization. We then developed a spatial polarimetric gray-level co-occurrence matrix (spGLCM) method to explore polarization-mediated textural differences in the bone collagen mesh. We used our spGLCM method to describe the collagen organizational differences between HPP and healthy bone along the polarimetric axis that may be caused by poorly aligned collagen molecules and a reduction in collagen density. Finally, we applied machine learning classifiers to predict bone disease state using pSHG imaging and spGLCM methods. Comparing random forest (RF) and XGBoost technique on spGLCM, we were able to accurately separate unknown images from the two groups with an averaged F1 score of 92.30%±3.11% by using RF. Our strategy could potentially allow for monitoring of therapeutic efficacy and disease progression in HPP, or even be extended to other collagen-related ailments or tissues.