Preparation and supervised segmentation of optical coherence tomography (OCT) images of the human uterotubal junction

The uterotubal junction (UTJ) is the muscular proximal segment of the fallopian tube that plays a critical role in embryo transport. Abnormal UTJ structure and function has been implicated in the pathogenesis of endometriosis and tubal factor infertility; however, this relationship remains poorly characterized. Our objective was to establish a framework for preprocessing of images to enable quantitative analysis of collagen in the UTJ using optical coherence tomography (OCT). UTJ tissue from 9 individuals undergoing hysterosalpingectomy were collected and longitudinally opened to expose the luminal surface. Non-overlapping volumetric OCT images (FOV: 3 x 1 x 2.5mm) were acquired along each UTJ segment using a benchtop OCT system. Images were preprocessed to enhance contrast of collagen fibers, suppress artifacts, and denoise and smooth features while preserving the continuity of collagen fibers. A supervised machine learning classifier (Trainable Weka Segmentation plugin, ImageJ) was iteratively trained and applied to each volume to generate a binary tissue mask. The mask was refined to exclude non-relevant areas and artifacts. An automated surface detection and smoothing algorithm was developed to remove overlying artifacts and nonrelevant structures from the volume. Our approach establishes a framework for the preparation of UTJ OCT volumes for downstream quantitative analysis of collagen fiber structure.