Screening of Wireless Capsule Endoscopy Image Sequence Based on Topic Model
Nong Guixian1,2, Pan Ning1,2, Lu Heng3, Hu Huaifei1,2, Liu Haihua1,2*
1(School of Biomedical Engineering, South-Central University for Nationalities University, Wuhan 430074, China) 2(Key Laboratory of Medical Information Analysis and Oncology Diagnosis, Wuhan 430074, China) 3(Department of Gastroenterology and Hepatology, Jinling Hospital, Medical School of Nanjing University, Nanjing 210002, China)
Abstract：Wireless capsule endoscopy (WCE), as a novel technology used to record images of the patient′s digestive tract, has greatly helped in the diagnosis of digestive tract diseases. However, during the detecting process, about 50,000-80,000 of images are produced for each patient, containing many disturbing images such as bubbles and impurities that greatly affect the efficiency of disease diagnosis. Most of the existing image screening methods, which are usually unstable and poorly generic, only have targeted on bubble images. Therefore, this paper proposed a topic model-based semantic analysis method to screen disturbing images from WCE sequences. The method firstly built an asymmetric autoencoder for image feature extraction, and used K-means algorithm to cluster the features of image patches in the training set to construct visual words; Secondly, the features of image patche in testing set were mapped into visual words to obtain the word frequency matrix of test images, resulting in the semantic representation of images based on visual words; Finally, the topic model was used to analyze the word frequency matrix and obtain the semantic classification of images. In this paper, the WCE dataset was obtained from 30 different patients and images in this dataset were annotated by a clinician with rich clinical experience. This dataset included 3 340 bubble images, 3 330 impurity images and 3 330 normal images, which were randomly divided into training and test sets in the ratio of 1∶1 for 10 times cross-validation. The experimental results showed that the proposed method effectively screened out the disturbing images, and the convolutional autoencoder based on the deep learning outperformed the traditional feature extraction method, obtaining 96.87% accuracy and effectively improving the efficiency of disease diagnosis.
农桂仙, 潘宁, 陆恒, 胡怀飞, 刘海华. 基于主题模型的胶囊内镜图像序列筛查[J]. 中国生物医学工程学报, 2022, 41(4): 412-419.
Nong Guixian, Pan Ning, Lu Heng, Hu Huaifei, Liu Haihua. Screening of Wireless Capsule Endoscopy Image Sequence Based on Topic Model. Chinese Journal of Biomedical Engineering, 2022, 41(4): 412-419.
[1] Iddan G, Meron G, Glukhovsky A, et al. Wireless capsule endoscopy[J]. Nature, 2000, 405(6785): 417-417. [2] Iakovidis DK, Koulaouzidis A. Software for enhanced video capsule endoscopy: challenges for essential progress[J]. Nature Reviews Gastroenterology & Hepatology, 2015, 12(3): 172-186. [3] 王晓艳, 王芬, 肖定华, 等. 无线胶囊内镜在消化道疾病诊断中的应用分析[J]. 中国内镜杂志, 2004, 10(12): 11-14. [4] Fu Y, Mandal M, Guo G. Bleeding region detection in WCE images based on color features and neural network[C]//2011 IEEE 54th International Midwest Symposium on Circuits and Systems (MWSCAS). Seoul: IEEE, 2011: 1-4. [5] 施慧英, 樊梦科, 王玮珺, 等. 基于深度卷积神经网络的小肠胶囊内镜智能辅助系统识别小肠出血[J]. 中华消化杂志, 2020, 40(11):763-767. [6] 范姗慧, 刘士臣, 曹鹗, 等. 无线胶囊内窥镜图像小肠息肉的自动识别[J]. 中国生物医学工程学报, 2019, 38(5):522-532. [7] Lan L, Ye C, Wang C, et al. Deep convolutional neural networks for WCE abnormality detection: CNN architecture, region proposal and transfer learning[J]. IEEE Access, 2019, 7: 30017-30032. [8] Yang W, Cao Y, Zhao Q, et al. Lesion classification of wireless capsule endoscopy images[C]//2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019). Venice: IEEE, 2019: 1238-1242. [9] Wang Q, Pan N, Xiong W, et al. Reduction of bubble-like frames using a RSS filter in wireless capsule endoscopy video[J]. Optics & Laser Technology, 2019, 110: 152-157. [10] Koulaouzidis A, Iakovidis DK, Karargyris A, et al. Optimizing lesion detection in small-bowel capsule endoscopy: from present problems to future solutions[J]. Expert Review of Gastroenterology & Hepatology, 2015, 9(2): 217-235. [11] Suman S, Hussin FA, Walter N, et al. Automatic detection and removal of bubble frames from wireless capsule endoscopy video sequences[C]//2016 6th International Conference on Intelligent and Advanced Systems (ICIAS). Kuala Lumpur: IEEE, 2016: 1-5. [12] Sun Z, Li B, Zhou R, et al. Removal of non-informative frames for wireless capsule endoscopy video segmentation[C]//2012 IEEE International Conference on Automation and Logistics. Zhengzhou: IEEE, 2012: 294-299. [13] de Iorio F, Radeva P, Vitria J, et al. Automatic detection of intestinal juices in wireless capsule video endoscopy[C]//18th International Conference on Pattern Recognition (ICPR′06). Hong Kong: IEEE, 2006, 4: 719-722. [14] Bashar MK, Mori K, Suenaga Y, et al. Detecting informative frames from wireless capsule endoscopic video using color and texture features[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2008: 603-610. [15] 刘小燕, 龚军辉, 李向东, 等. 基于 SURF 算法的胶囊内镜冗余图像快速筛除[J]. 中国生物医学工程学报, 2016, 35(3): 264-271. [16] Sivic J, Russell BC, Efros AA, et al. Discovering objects and their location in images[C]//Tenth IEEE International Conference on Computer Vision (ICCV′05) Volume 1. Beijing: IEEE, 2005, 1: 370-377. [17] Bosch A, Zisserman A, Munoz X. Scene classification via pLSA[C]//European conference on computer vision. Berlin: Springer, 2006: 517-530. [18] 张瑞杰, 李弼程, 魏福山. 基于多尺度上下文语义信息的图像场景分类算法[J]. 电子学报, 2014, 42(4):646-652. [19] Fei-Fei L, Perona P. A bayesian hierarchical model for learning natural scene categories[C]//2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR′05). San Diego: IEEE, 2005, 2: 524-531. [20] 赵仲秋, 季海峰, 高隽,等. 基于稀疏编码多尺度空间潜在语义分析的图像分类[J]. 计算机学报, 2014(6):1251-1260. [21] Vayansky I, Kumar SAP. A review of topic modeling methods[J]. Information Systems, 2020, 94: 101582. [22] Hofmann T. Probabilistic latent semantic analysis[DB/OL]. https://arxiv.org/abs/1301.6705,2013-01-23/2021-10-01. [23] Blei DM, Ng AY, Jordan MI. Latent dirichlet allocation[J]. Journal of Machine Learning Research, 2003, 3: 993-1022. [24] Biniaz A, Zoroofi RA, Sohrabi MR. Automatic reduction of wireless capsule endoscopy reviewing time based on factorization analysis[J]. Biomedical Signal Processing and Control, 2020, 59: 101897. [25] Shen Y, Guturu P, Buckles B P. Wireless capsule endoscopy video segmentation using an unsupervised learning approach based on probabilistic latent semantic analysis with scale invariant features[J]. IEEE Transactions on Information Technology in Biomedicine, 2011, 16(1): 98-105. [26] Yuan Y, Yao X, Han J, et al. Discriminative joint-feature topic model with dual constraints for WCE classification[J]. IEEE Transactions on Cybernetics, 2017, 48(7): 2074-2085. [27] LeCun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521(7553): 436-444. [28] 袁非牛, 章琳, 史劲亭,等. 自编码神经网络理论及应用综述[J]. 计算机学报, 2019, 42(1):203-230. [29] Kingma DP, Ba J. Adam: a method for stochastic optimization[DB/OL].https://arxiv.org/abs/1412.6980,2017-01-30/2021-10-01. [30] MacQueen J. Some methods for classification and analysis of multivariate observations[C]//Proceedings of the 5th Berkeley Symposium on Mathematical Statistics and Probability. Berkeley: University of California Press, 1967, 1(14): 281-297.
