Automatic Epidermis Segmentation in Skin Fluorescence Images Based on Multistage Active Contour Models
Journal
Proceedings - 2024 2nd International Conference on Computer Graphics and Image Processing, CGIP 2024
Start Page
1
End Page
6
ISBN (of the container)
9798350374186
ISBN
9798350374186
Date Issued
2024-01-12
Author(s)
DOI
10.1109/CGIP62525.2024.00009
Abstract
Small fiber neuropathy is one of the common manifestations of peripheral nerve neuropathy. Quantification of intraepidermal nerve fiber density (IENFd) in the epidermis, which is the most superficial layer of the skin, plays an important role in examining the nerve pathology. An essential task is to segment the epidermis region in skin fluorescence images for subsequent investigation. The greater use of computer-aided technology in digital image analysis and modeling possesses particular promise. Due to the lacking of available methods, this paper proposes an automatic epidermis segmentation framework based on multistage active contour models. The color fluorescence image is first decomposed into three channels according to the CIEL*a*b∗ color model. A series of image processing techniques are then developed to adaptively acquire the initial contour for the subsequent curve evolution. On account of the huge dimension of the skin fluorescence images, a three-stage active contour segmentation model is proposed to facilitate the task while improving the accuracy and saving the computation time. Experiments on 50 in-house epidermis images demonstrated the advantages of the proposed multistage segmentation scheme. The average Dice similarity score was 81.55%, which outperformed other competing methods. With the deficiency of a large epidermis image dataset for deep learning-based method development, the introduced segmentation model is potential in providing an appropriate tool for small fiber neuropathy research.
Event(s)
2nd International Conference on Computer Graphics and Image Processing, CGIP 2024, Kyoto 12 January 2024 through 15 January 2024, Code 199665
Subjects
active contour modeling
epidermis tissue
image segmentation
Publisher
IEEE
Type
conference paper