«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

32

期数:

2022年04期

页码:

176-180

栏目:

应用前沿与综合

出版日期:

2022-04-10

文章信息/Info

Title:

Breast Cancer Histopathological Image Classification Based on Multi-scale and Multi-gamut Feature Fusion

文章编号:

1673-629X(2022)04-0176-05

作者:

张 莉¹ ; 张 成¹ ; 郝 岩² ; 程 蓉¹ ; 白艳萍¹

1. 中北大学 理学院,山西 太原 030051;

2. 中北大学 信息与通信工程学院,山西 太原 030051

Author(s):

ZHANG Li1 ; ZHANG Cheng1 ; HAO Yan2 ; CHENG Rong1 ; BAI Yan-ping1

1. School of Science,North University of China,Taiyuan 030051,China;

2. School of Information and Communication Engineering,North University of China,Taiyuan 030051,China

关键词:

乳腺癌; 颜色矩; 灰度共生矩阵; Haar 小波; 支持向量机; 多数投票策略

Keywords:

breast cancer; color moments; gray level co-occurrence matrix; Haar wavelet; support vector machine; majority voting strategy

分类号:

TP391.4

DOI:

10. 3969 / j. issn. 1673-629X. 2022. 04. 030

摘要:

乳腺癌是世界上女性发病率最高的癌症,而组织病理图像是鉴定乳腺癌的“黄金标准” 。 为了实现对乳腺癌组织病理图像的精确分类,提出了一种基于多尺度多色域特征融合的乳腺癌组织病理图像分类方法。 此方法能实现良、恶性病理图像的有效分类。 首先进行多色域特征提取,将病理图像从 RGB 空间转换到 HSV 空间,分别提取 H、S、V 三个色彩分量的 9 维颜色矩特征和 24 维灰度共生矩阵特征( GLCM) ;其次进行多尺度特征提取, 利用 Haar 两层小波分解提取病理图像的高频分量( 水平、垂直、对角) ,共得高频分量的 48 维灰度共生矩阵特征。 将最终形成的 81 维特征向量输入到不同训练集训练所生成的 7 类支持向量机( SVM) 中进行分类,将分类结果采取多数投票策略,获得最终识别准确率。 通过BreaKHis 公开数据集的实验表明,4 个放大倍数图像的分类准确率分别达到约 95. 31% 、94. 34% 、93. 07% 和 91. 94% 。

Abstract:

Breast cancer is the most frequent cancer in women in the world,and his to pathological images are the " gold standard" for identifying breast cancer. To achieve precise classification of breast cancer his to pathological images, a method for breast cancer image classification based on multi-scale multi - color feature fusion is proposed. Such method can effectively classify benign and malignant pathological images. First,multi-color domain feature extraction was performed to transform the pathological images from RGB space to HSV space,and 9-dimensional color moment features and 24 dimensional gray level co-occurrence matrix features ( GLCM) of the three color components of H, S, and V were extracted, respectively. Next, multi - scale feature extraction was performed, high - frequency components ( horizontal, vertical, diagonal ) of breast cancer pathological images were extracted using Haar two - layer wave let decomposition,and 48 dimensional gray level co-occurrence matrix features with high-frequency components were obtained. The final formed 81 dimensional feature vector was input into the 7 - class support vector machine ( SVM) generated by training with different training sets for classification, and the classification results were taken a majority voting strategy to obtain the final identification accuracy. The classification accuracy of the four magnification images reached approximately 95. 31% ,94. 34% ,93. 07% ,91. 94% ,respectively,as determined by the results of experiments with BreaKH is publicly available datasets.

常用功能

更新日期/Last Update: 2022-04-10