Computational Imaging Approaches For Cancer Stromal Biomarker Characterization PDF Download

Collagen forms the structural network of the extracellular matrix (ECM) in biological tissues and is the most abundant protein in vertebrates. The amount, distribution, and structural organization of fibrillar collagen are all important factors underlying the properties of tissues and play an integral role in many diseases, including cancer. There have been many studies using collagen as a biomarker in wound healing, aging, and other pathologies including fibrosis, atherosclerosis and diabetes. Most of these findings are due to advent of Second Harmonic Generation (SHG) microscopy, a type of laser scanning two-photon microscopy specific to non-cetnrosymmetric structures such as collagen molecule. However, due to the high cost, size and the need for expert operator this modality has never been clinically used. Imaging cost and lack of proper image analysis tools have been major barriers towards interrogating these stromal biomarkers and finding new insights in clinical research and diagnosis. Here we present some answers to these questions by introducing CurveAlign, a collagen analysis software package that has been widely used for quantitation of collagen images that has led to new insights regarding stromal reorganization during disease progression from wound healing to breast, pancreatic, renal cell and other types of cancer. To overcome the impracticality of using SHG microscopy in clinical pathology we have introduced and explored a variety of polarization based, fluorescence and phase imaging methods that are all clinically feasible and could be setup on pathologist microscope. However, to further facilitate the collagen imaging on H & E stained samples that have been the gold standard in diagnostic pathology we have proposed a method to computationally synthesize SHG images of collagen from bright field images of these samples as the pathologist see using microscope eyepieces by using convolutional neural networks. To further prove applicability of stromal biomarkers in pathological diagnosis we have tested whole tissue (cell and stroma) evaluation using deep learning methods to improve the diagnosis and prognosis of pancreatic cancer, which is one of the deadliest types cancer with a unique stromal response during cancer progression.