AUTOMATED DETECTION OF COLLAGEN BUNDLES ON SECOND HARMONIC GENERATION IMAGES

Abstract

Collagen is one of the most abundant proteins
 in the body. It is essential for the structure, functionality, and
 strength of the connective tissue such as skin, bone, tendon,
 and cornea. It is known that a change in the arrangement
 or morphology of these fibrillar structures relates to multiple
 dysfunctions including corneal diseases and various cancer types.
 Due to their critical roles in wide-range abnormalities, there
 is an increasing interest in the pattern analysis of collagen
 arrangements. In recent years, Second Harmonic Generation
 (SHG) microscopy is proven to be an efficient imaging modality
 for visualizing unstained collagen fibrils. There are plenty of
 studies in the literature on the analysis of collagen distribution
 in SHG images. However, the majority of these methods are
 limited to detecting simple, statistical and non-local properties
 such as pixel intensity and orientation variance. There is a
 need for a method to detect the local structural properties of
 collagen bundles. This paper is to introduce an automated method
 to detect collagen bundles in 3-dimensional SHG microscopy
 images. The origin of the proposed method is based on multiscale
 directional representation systems. The proposed method detects
 the collagen bundles by measuring the dominant orientation
 of local regions and an orientation-based connected component
 analysis. Through more local analysis and the detection of
 collagen bundles separately, the proposed method would lead to
 the extraction of more detailed structural information on collagen
 bundle distribution.