Related Experiment Video
Updated: Sep 30, 2025
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
2.7K
Study on TCM Tongue Image Segmentation Model Based on Convolutional Neural Network Fused with Superpixel
Han Zhang1, Rongrong Jiang2, Tao Yang1,3
1School of Artificial Intelligence and Information Technology, Nanjing University of Chinese Medicine, Nanjing, China.
Evidence-Based Complementary and Alternative Medicine : Ecam
|March 18, 2022
Summary
This study introduces SpurNet, a novel deep learning model that improves tongue image segmentation by fusing convolutional neural networks with superpixel segmentation. SpurNet enhances accuracy in Traditional Chinese Medicine tongue analysis, particularly for challenging images with similar tongue and background colors.
Area of Science:
- Computer Vision
- Medical Imaging
- Traditional Chinese Medicine (TCM)
Background:
- Deep learning methods enhance tongue segmentation in TCM, but struggle with images where tongue color closely matches surrounding areas, leading to imprecise contours.
- Coarse tongue segmentation can negatively impact subsequent analyses in TCM tongue processing.
- Existing deep learning models like FCN and DeepLab face limitations in accurately segmenting tongue images with ambiguous boundaries.
Purpose of the Study:
- To propose a novel tongue image segmentation model, SpurNet, that overcomes the limitations of existing deep learning methods.
- To improve the accuracy of tongue contour segmentation, especially in cases with low contrast between the tongue and surrounding regions.
- To provide a more robust tool for TCM tongue image analysis.
Main Methods:
- Developed SpurNet, a model fusing a convolutional neural network (CNN) with superpixel segmentation.
- Utilized ResNet18's residual structure for feature extraction in the encoding path (UrNet).
- Integrated superpixel segmentation with UrNet to form the complete SpurNet model, tested on 367 manually labeled tongue images.
Main Results:
- SpurNet achieved high performance metrics: Pixel Accuracy (PA) of 0.9145 ± 0.0043, Mean Pixel Accuracy (MPA) of 0.9168 ± 0.0048, Mean Intersection over Union (MIoU) of 0.8417 ± 0.0072, and Frequency Weighted Intersection over Union (FWIoU) of 0.8454 ± 0.0072.
- SpurNet outperformed FCN and DeepLab, increasing PA by 1.91%-3.17%, MPA by 1.38%-2.61%, MIoU by 3.09%-5.07%, and FWIoU by 3.11%-5.08%.
- Compared to its initial stage (UrNet), SpurNet further improved PA, MPA, MIoU, and FWIoU by 0.15%, 0.09%, 0.24%, and 0.24%, respectively.
Conclusions:
- The SpurNet model effectively enhances tongue image segmentation accuracy by integrating superpixel segmentation with CNNs.
- SpurNet accurately processes tongue margins and resolves over-segmentation and under-segmentation issues, outperforming previous methods.
- This novel approach offers a valuable reference for advanced research in TCM tongue image analysis.
