The morphological characteristics of urine bubbles are an important objective basis for identifying Heyi syndrome and other syndromes in Mongolian medicine urine diagnosis. However, traditional Mongolian medicine urine diagnosis relies on visual judgment of the "three times and nine laws" features, which suffers from inaccuracies due to human errors and lacks a standardized dataset. To address these issues, a Mongolian medicine urine diagnosis dataset named MUB-Seg, containing 2 400 samples, was established. This dataset included five features from the "three times and nine laws" of Mongolian medicine urine diagnosis: color, bubbles, transparency, sediment, and floating matter across the warm, hot, and cold phases. Based on VMUNet, a multi-scale feature fusion network named MAU-Net was constructed to improve bubble extraction accuracy. Specifically, an atrous spatial pyramid pooling module was combined with U-Net cascaded decoding to enhance the model's efficiency in capturing global image information and local details. The loss function jointly optimized binary cross-entropy loss and Dice loss to balance small-target recall and boundary alignment accuracy, avoiding the bias of a single loss function. Experimental results on the self-built MUB-Seg dataset show that MAU-Net combined with the BCE-Dice composite loss achieves an IoU of 72.88%, an F1 score of 84.31%, and a sensitivity of 83.75%. It outperforms PSPNet, U2-Net, U-Net, and VMUNet in edge detail recovery, small bubble detection, and complex background suppression, improving by 3.11%, 2.11%, and 2.51%, respectively, compared with the suboptimal model PSPNet. Thus, the strategy of multi-scale feature fusion and loss function co-optimization effectively enhances the sensitivity of MAU-Net to local features and structural boundaries of urine bubble images, providing technical support for the objective identification of Heyi syndrome while also demonstrating the feasibility of deep learning in ethnic medical image analysis.
LUOH W, MAS L, WUD Y, et al. Mumford-Shah segmentation for microscopic image of the urinary sediment[C]//2007 1st International Conference on Bioinformatics and Biomedical Engineering. Wuhan: IEEE, 2007: 861-863.
[2]
MUMFORDD, SHAHJ. Optimal approximations by piecewise smooth functions and associated variational problems[J]. Communications on Pure and Applied Mathematics, 1989, 42(5): 577-685.
[3]
JIANGX, NIES D. Urine sediment image segmentation based on level set and Mumford-Shah model[C]//2007 1st International Conference on Bioinformatics and Biomedical Engineering. Wuhan: IEEE, 2007: 1028-1030.
[4]
OSHERS, SETHIANJ A. Fronts propagating with curvature-dependent speed: Algorithms based on Hamilton-Jacobi formulations[J]. Journal of Computational Physics, 1988, 79(1): 12-49.
[5]
LIY M, ZENGX P. A new strategy for urinary sediment segmentation based on wavelet, morphology and combination method[J]. Computer Methods and Programs in Biomedicine, 2006, 84(2-3): 162-173.
[6]
ZHANGZ C, XIAS R, DUANH L. Cellular neural network based urinary image segmentation[C]//Third International Conference on Natural Computation (ICNC 2007). Haikou: IEEE, 2007: 285-289.
[7]
ZHANGS, WANGJ H, ZHAOS G, et al. Urinary sediment images segmentation based on efficient Gabor filters[C]//2007 IEEE/ICME International Conference on Complex Medical Engineering. Beijing: IEEE, 2007: 812-815.
[8]
LIC Y, FANGB, WANGY, et al. Automatic detecting and recognition of casts in urine sediment images[C]//2009 International Conference on Wavelet Analysis and Pattern Recognition. Baoding: IEEE, 2009: 26-31.
[9]
WUY H, KANGM. A urinary sediment segmentation algorithm of cast and epithelial cells[J]. Applied Mechanics and Materials, 2013(457-458): 1046-1049.
[10]
ZHANGS, QIJ L, QIG J. Urinary sediment overlapping cells image segmentation based on combination strategy[C]//2008 International Symposium on Computational Intelligence and Design. Wuhan: IEEE, 2008: 3-7.
[11]
SUL C, JIANGS, WANGC L. A segmentation method based on standard difference gradient with dual-threshold for urinary sediment visible components[C]//2017 4th International Conference on Information Science and Control Engineering (ICISCE). Changsha: IEEE, 2017: 94-98.
[12]
ARBELÁEZP, MAIREM, FOWLKESC, et al. Contour detection and hierarchical image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(5): 898-916.
[13]
FAND P, JIG P, ZHOUT, et al. PraNet: Parallel reverse attention network for polyp segmentation[M]//Medical Image Computing and Computer Assisted Intervention-MICCAI 2020. Lima: Springer, 2020: 263-273.
[14]
BUIN T, HOANGD H, NGUYENQ T, et al. MEGANet: Multi-scale edge-guided attention network for weak boundary polyp segmentation[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. Waikoloa: IEEE, 2024: 7985-7994.
[15]
KRIZHEVSKYA, SUTSKEVERI, HINTONG E. Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
[16]
SIMONYANK, ZISSERMANA. Very deep convolutional networks for large-scale image recognition[J]. arXiv preprint arXiv, 2014:1409.1556.
[17]
SZEGEDYC, LIUW, JIAY Q, et al. Going deeper with convolutions[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston: IEEE, 2015: 1-9.
[18]
HEK, ZHANGX, RENS, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 770-778.
[19]
HUANGG, LIUZ, VAN DER MAATENL, et al. Densely connected convolutional networks[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu: IEEE, 2017: 2261-2269.
[20]
LONGJ, SHELHAMERE, DARRELLT. Fully convolutional networks for semantic segmentation[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston: IEEE, 2015: 3431-3440.
[21]
RONNEBERGERO, FISCHERP, BROXT. U-Net: Convolutional networks for biomedical image segmentation[M]//Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015. Munich: Springer, 2015: 234-241.
[22]
RUANJ, LIJ, XIANGS. VM-UNet: Vision Mamba UNet for medical image segmentation[J]. arXiv preprint arXiv, 2024: 2402.02491.
[23]
CHENL C, ZHUY K, PAPANDREOUG, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[M]//Computer Vision-ECCV 2018. Munich: Springer, 2018: 801-818.
[24]
QINX B, ZHANGZ C, HUANGC Y, et al. U2-Net: Going deeper with nested U-structure for salient object detection[J]. Pattern Recognition, 2020, 106: 107404.