SHPNeXt: A Novel Method of Multi-Scale and Variable Resolution AI-Based Tongue Image Segmentation

C. Y. Peng, Zhijun Gao, Jinhuan Wang, Xin Yue, Yi Li, Lili Sun, Ying Sun, Fanyi DuIEEE Access 2025 · DOI

Abstract

In the domain of Traditional Chinese Medicine, accurately segmenting tongue images is fundamental for computer-assisted diagnosis. Yet, current models often falter with images of diverse scales and clarity, impeding their widespread application. To address this challenge, we propose SHPNeXt, an innovative network designed to accurately segment tongue images across different scales and resolutions. This model blends PoolFormer and Hire-MLP to adeptly discern both overarching and nuanced details, ensuring accurate segmentation across varying tongue image sizes. Furthermore, SHPNeXt’s precision was further enhanced by integrating a Nuclear-Norm Non-negative Matrix Factorization (NMF) approach, which robustly counters noise in lower quality images. Experiments on three benchmark datasets demonstrate SHPNeXt’s superior performance, achieving mean Intersection over Union (mIoU) scores of 99.64%, 97.05%, and 96.82%. Balancing efficiency and accuracy, SHPNeXt’s architecture comprises 5.984 million parameters and operates at 1.22 GFLOPs, rendering it an exceptionally effective tool for real-world tongue diagnosis in TCM. The code has been released on github: (<uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/Kuanzhaipcx/SHPNeXt.git</uri>).
BibTeX
@inproceedings{peng2025shpnext,
  author = {C. Y. Peng and Zhijun Gao and Jinhuan Wang and Xin Yue and Yi Li and Lili Sun and Ying Sun and Fanyi Du},
  title = {{SHPNeXt: A Novel Method of Multi-Scale and Variable Resolution AI-Based Tongue Image Segmentation}},
  booktitle = {IEEE Access},
  year = {2025},
  doi = {10.1109/access.2025.3554487},
}

Loading related content...

Figure