Differentiating morphea from lichen Sclerosus by using multiphoton microscopy combined with U-Net model for elastic fiber segmentation

Abstract

This paper describes a methodology to differentiate morphea from lichen sclerosus based on examination with multiphoton microscopy (MPM) composed of two-photon excited fluorescence (TPEF) and second harmonic generation (SHG). Subcellular-resolution images were acquired by MPM from unstained lesion tissues then process spectral analysis to quantify the TPEF and SHG signals. Moreover, U-Net was employed to segment elastic fiber in TPEF images to combine with collagen fiber in SHG images for precise fiber quantification. Predictions of segmentation showed excellent performance on several evaluation indicators. The mIoU, mPA, and F1 score reach 0.8516, 0.9281, and 0.941. The quantitative analysis demonstrated the increase of collagen fibers in morphea compared to that in lichen sclerosus cases. Meanwhile, the great diminution of elastic fiber in the dermis of lichen sclerosus was depicted based on MPM imaging. Thus, MPM was comparable to the histopathological examination and our experimental results accurately distinguish between morphea and lichen sclerosus.

Keywords: U-net; lichen sclerosus; morphea; multiphoton microscopy; segmentation.