A lightweight PCT-Net for segmenting neural fibers in low-quality CCM images

Le Liu, Chen Tang, Zhong Ji, Jia He Yang, Jia Yi Jiang, Zhen Kun Lei

Abstract: Highlights•A lightweight Position Channel Transformer Network (PCT-Net) was proposed for segmenting slender neural fibers in low-quality CCM images with speckle noise and uneven lighting.•A hybrid loss function was proposed, which combines MS-SSIM loss, BCE loss, and polyfocal loss to train PCT-Net.•The problem of imbalanced label categories in the dataset was solved by using morphological unfolding operations on structural elements to expand fiber labels.•The proposed method could protect the structural information of nerve fibers and improve segmentation accuracy, and be used for other low-quality CCM image datasets.