TY - JOUR
T1 - A novel framework for tongue feature extraction framework based on sublingual vein segmentation
AU - Wan, Xiaohua
AU - Hu, Yulong
AU - Qiu, Dehui
AU - Zhang, Juan
AU - Wang, Xiaotong
AU - Zhang, Fa
AU - Hu, Bin
N1 - Publisher Copyright:
© 2002-2011 IEEE.
PY - 2024
Y1 - 2024
N2 - The features of the sublingual veins, including swelling, varicose patterns, and cyanosis, are pivotal in differentiating symptoms and selecting treatments in Traditional Chinese Medicine (TCM) tongue diagnosis. These features serve as a crucial reflection of the human blood circulation status. Nevertheless, the automatic and precise extraction of sublingual vein features remains a formidable challenge, constrained by the scarcity of datasets for sublingual images and the interference of noise from non-tongue and non-sublingual vein elements. In this paper, we present an innovative tongue feature extraction method that relies on focusing specifically on segmenting the sublingual vein rather than the entire tongue base. To achieve this, we have developed a sublingual vein segmentation framework utilizing a Polyp-PVT network, effectively eliminating noise from the surrounding regions of the sublingual vein. Furthermore, we pioneer the utilization of a transformer-based approach, such as the Swin-Transformer network, to extract sublingual vein features, leveraging the remarkable capabilities of transformer networks. To complement our methodology, we have constructed a comprehensive dataset of sublingual vein images, facilitating the segmentation and classification of sublingual veins. Experimental results have demonstrated that our tongue feature extraction method, coupled with sublingual vein segmentation, significantly outperforms existing tongue feature extraction techniques.
AB - The features of the sublingual veins, including swelling, varicose patterns, and cyanosis, are pivotal in differentiating symptoms and selecting treatments in Traditional Chinese Medicine (TCM) tongue diagnosis. These features serve as a crucial reflection of the human blood circulation status. Nevertheless, the automatic and precise extraction of sublingual vein features remains a formidable challenge, constrained by the scarcity of datasets for sublingual images and the interference of noise from non-tongue and non-sublingual vein elements. In this paper, we present an innovative tongue feature extraction method that relies on focusing specifically on segmenting the sublingual vein rather than the entire tongue base. To achieve this, we have developed a sublingual vein segmentation framework utilizing a Polyp-PVT network, effectively eliminating noise from the surrounding regions of the sublingual vein. Furthermore, we pioneer the utilization of a transformer-based approach, such as the Swin-Transformer network, to extract sublingual vein features, leveraging the remarkable capabilities of transformer networks. To complement our methodology, we have constructed a comprehensive dataset of sublingual vein images, facilitating the segmentation and classification of sublingual veins. Experimental results have demonstrated that our tongue feature extraction method, coupled with sublingual vein segmentation, significantly outperforms existing tongue feature extraction techniques.
KW - Traditional Chinese Medicine
KW - sublingual vein
KW - tongue diagnosis
KW - tongue feature extraction
KW - tongue image
UR - http://www.scopus.com/inward/record.url?scp=85204702342&partnerID=8YFLogxK
U2 - 10.1109/TNB.2024.3462461
DO - 10.1109/TNB.2024.3462461
M3 - Article
AN - SCOPUS:85204702342
SN - 1536-1241
JO - IEEE Transactions on Nanobioscience
JF - IEEE Transactions on Nanobioscience
ER -