基于K-means与宽度学习的肺炎图像分类算法  

Pneumonia image classification algorithm based on K-means and broad learning

作　　者：程立英[1,2] 谷利茹 晏旻 管文印 王晓伟 张志美 CHENG Liying;GU Liru;YAN Min;GUAN Wenyin;WANG Xiaowei;ZHANG Zhimei(College of Physical Science and Technology,Shenyang Normal University,Shenyang 110034,China;Key Laboratory of Intelligent Computing in Medical Image,Ministry of Education,Shenyang 110003,China)

机构地区：[1]沈阳师范大学物理科学与技术学院,沈阳110034 [2]医学影像智能计算教育部重点实验室,沈阳110003

出　　处：《沈阳师范大学学报（自然科学版）》2024年第4期334-339,共6页Journal of Shenyang Normal University:Natural Science Edition

基　　金：教育部产学合作协同育人项目(230701160110601)。

摘　　要：随着人们日常生活中肺部疾病风险的增加,肺部病变筛查变得至关重要。通过CT图像快速辅助诊断肺炎可以有效遏制病情。针对现有的肺部CT图像辅助诊断方法存在数据标记量大、训练数据耗时长以及对医疗设备计算量和内存要求高等问题,提出基于K-means与宽度学习的肺炎图像分类算法。该算法引入K-means使宽度学习系统更好地提取肺部CT图像特征,缓解随机获得节点权值的性能局限,建立与典型特征学习相关的宽度学习模型,并将算法针对公开数据集进行相关实验。实验结果表明,该模型较深度学习模型的计算量大大减小,在训练速度方面有明显优势,同时也保证了较好的分类结果,极大地降低了诊断时间;在数据有限的情况下,改进后的方法与现有主流方法相比获得了更加精确的肺炎诊断结果,提出的算法更适于嵌入医学设备等资源有限的硬件系统中。As the risk of lung diseases increases in people′s daily lives,screening for lung lesions has become critical.The rapid auxiliary diagnosis of pneumonia through CT images can effectively curb the disease.In view of the existing lung CT image-assisted diagnosis methods have problems such as large amounts of data labeling,long training data consumption,and high computational and memory requirements for medical equipment,this paper proposes a pneumonia image classification algorithm based on K-means and Broad learning.This algorithm introduces K-means to enable the broad learning system to better extract lung CT image features,alleviate the performance limitations of randomly obtaining node weights,establish a broad learning model related to typical feature learning,and target the algorithm to public data sets Related experiments were conducted.Experimental results show that this model requires significantly less calculations than the deep learning model and has obvious advantages in training speed.It also ensures better classification results and greatly reduces the diagnosis time;in the case of limited data,The improved network achieves more accurate pneumonia diagnosis results compared with existing mainstream methods.Therefore,the algorithm proposed in this article is more suitable for embedding into resource-limited hardware systems such as medical equipment.

关 键 词：肺炎诊断 CT图像 K-MEANS 宽度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]