基于纹理和统计图像特征集成的计算机辅助乳腺癌检测

167 浏览量 2024-07-18 21:17:17 上传评论收藏 14.3MB PDF 举报

### 基于纹理和统计图像特征集成的计算机辅助乳腺癌检测 #### 一、引言乳腺癌作为全球范围内最普遍的恶性肿瘤之一，其早期诊断与治疗对于提高患者的生存率至关重要。根据世界卫生组织的数据，乳腺癌的年死亡人数超过了50万，并在2020年成为最常见的癌症类型。尽管这种疾病极为致命，但若能早期发现并及时治疗，患者的存活率将大幅提高。因此，开发高效的辅助诊断技术显得尤为重要。 #### 二、研究背景及目的本文探讨了一种新型的计算机辅助乳腺癌检测方法，该方法通过集成纹理和统计图像特征来提高检测的准确性和可靠性。传统的乳腺癌检测通常依赖于医生对组织病理学图像的人工分析，这种方法不仅耗时费力，而且容易受到主观因素的影响。为了解决这些问题，本研究提出了一种基于机器学习的方法，旨在利用计算机自动识别和分类乳腺癌组织图像。 #### 三、关键技术点 1. **纹理特征提取**：纹理是图像的重要组成部分，对于区分不同类型的组织具有重要意义。本研究采用了多种纹理特征提取方法，如灰度共生矩阵(GLCM)、局部二值模式(LBP)等，这些方法能够捕捉到图像中的细节变化，从而为后续的分类提供关键信息。 2. **统计特征提取**：除了纹理特征外，研究还考虑了统计特征的重要性。这些特征包括均值、标准差、熵等，它们能够反映图像的全局特性，有助于增强模型的识别能力。 3. **特征集成**：为了充分利用纹理和统计特征的优势，本研究采用了一种特征集成的方法。通过组合不同的特征集，可以更全面地描述图像的信息，进而提高检测的准确性。 4. **机器学习算法**：在特征提取之后，选择了多种机器学习算法进行训练和测试，包括支持向量机(SVM)、随机森林(RF)等。这些算法能够从数据中学习到潜在的规律，从而实现对乳腺癌组织的有效分类。 #### 四、实验设计与结果 1. **数据集构建**：为了验证所提出方法的有效性，本研究构建了一个包含大量乳腺癌组织图像的数据集。这些图像经过专业病理学家的标注，确保了数据的质量和准确性。 2. **特征选择与优化**：通过对多种特征进行对比分析，选择了最优的特征组合用于模型训练。此外，还进行了参数调优，以进一步提升模型的性能。 3. **模型评估**：最终的模型在独立的测试集上进行了评估，结果显示，集成纹理和统计特征的方法相较于单一特征方法，在识别精度上有显著提高。 #### 五、结论与展望基于纹理和统计图像特征集成的计算机辅助乳腺癌检测方法能够有效提高诊断的准确性。这种方法不仅可以减轻医生的工作负担，还能帮助更多患者尽早获得合适的治疗方案。未来的研究将进一步探索深度学习等先进技术的应用，以期在提高诊断效率的同时，降低误诊率，为临床实践提供更加有力的支持。

资源推荐

资源详情

资源评论

sensors

Article

Computer Aided Breast Cancer Detection Using Ensembling of

Texture and Statistical Image Features

Soumya Deep Roy

, Soham Das

, Devroop Kar

, Friedhelm Schwenker

* and Ram Sarkar



 

Citation: Roy, S.D.; Das, S.; Kar, D.;

Schwenker, F.; Sarkar, R. Computer

Aided Breast Cancer Detection Using

Ensembling of Texture and Statistical

Image Features. Sensors 2021, 21, 3628.

https://doi.org/10.3390/s21113628

Academic Editor: Sheryl Berlin

Brahnam

Received: 12 April 2021

Accepted: 14 May 2021

Published: 23 May 2021

Publisher’s Note: MDPI stays neutral

with regard to jurisdictional claims in

published maps and institutional afﬁl-

iations.

 2021 by the authors.

Licensee MDPI, Basel, Switzerland.

This article is an open access article

distributed under the terms and

conditions of the Creative Commons

Attribution (CC BY) license (https://

creativecommons.org/licenses/by/

4.0/).

Department of Metallurgical and Material Engineering, Jadavpur University, Kolkata 700032, India;

[email protected] (S.D.R.); [email protected] (S.D.)

Department of Computer Science and Engineering, Jadavpur University, Kolkata 700032, India;

kardevr[email protected] (D.K.); [email protected] (R.S.)

Institute of Neural Information Processing, Ulm University, 89081 Ulm, Germany

* Correspondence: [email protected]

Abstract:

Breast cancer, like most forms of cancer, is a fatal disease that claims more than half a

million lives every year. In 2020, breast cancer overtook lung cancer as the most commonly diagnosed

form of cancer. Though extremely deadly, the survival rate and longevity increase substantially with

early detection and diagnosis. The treatment protocol also varies with the stage of breast cancer.

Diagnosis is typically done using histopathological slides from which it is possible to determine

whether the tissue is in the Ductal Carcinoma In Situ (DCIS) stage, in which the cancerous cells have

not spread into the encompassing breast tissue, or in the Invasive Ductal Carcinoma (IDC) stage,

wherein the cells have penetrated into the neighboring tissues. IDC detection is extremely time-

consuming and challenging for physicians. Hence, this can be modeled as an image classiﬁcation task

where pattern recognition and machine learning can be used to aid doctors and medical practitioners

in making such crucial decisions. In the present paper, we use an IDC Breast Cancer dataset

that contains 277,524 images (with 78,786 IDC positive images and 198,738 IDC negative images)

to classify the images into IDC(+) and IDC(-). To that end, we use feature extractors, including

textural features, such as SIFT, SURF and ORB, and statistical features, such as Haralick texture

features. These features are then combined to yield a dataset of 782 features. These features are

ensembled by stacking using various Machine Learning classiﬁers, such as Random Forest, Extra

Trees, XGBoost, AdaBoost, CatBoost and Multi Layer Perceptron followed by feature selection

using Pearson Correlation Coefﬁcient to yield a dataset with four features that are then used for

classiﬁcation. From our experimental results, we found that CatBoost yielded the highest accuracy

(92.55%), which is at par with other state-of-the-art results—most of which employ Deep Learning

architectures. The source code is available in the GitHub repository.

Keywords: breast cancer; IDC; machine learning; ensemble learning; feature selection

1. Introduction

With the widespread digitization of health records, computer aided disease detection

(CADD) systems that employ data mining and Machine Learning (ML) techniques have

become increasingly commonplace. Considering the monstrosity of the disease, it comes

as little surprise that the earliest efforts in CADD [

] started with mammography for the

detection of breast cancer and were later extended to other types of cancer as well. Breast

cancer is a condition in which the cells in the breast proliferate uncontrollably.

According to global cancer statistics [

], breast cancer in women has overtaken lung

cancer as the most diagnosed cancer in the world, comprising almost 12% of the total

instances of cancer worldwide. As is the case in most types of cancer, the treatment of

breast cancer is greatly aided by early detection but involves surgical or intensive medical

procedures if not diagnosed early.

Sensors 2021, 21, 3628. https://doi.org/10.3390/s21113628 https://www.mdpi.com/journal/sensors

Sensors 2021, 21, 3628 2 of 17

Since treatment depends on the stage of cancer, one of the preliminary tasks of any

pathologist involves a visual analysis of a histopathological slide stained with hematoxylin

and eosin (H&E). It is possible to determine whether the tissue is in the Ductal Carcinoma

In Situ (DCIS) stage or in the Invasive Ductal Carcinoma (IDC) stage from the slide. DCIS

is a non-invasive cancer where cancerous cells have been found in the lining of the breast

milk duct but have not spread into the surrounding breast tissue. However, in IDC, the

abnormal cancer cells that began forming in the milk ducts spread beyond the ducts into

other parts of the breast tissue. IDC is the most prevalent breast cancer type, responsible for

about 70–80% of all breast cancer diagnoses. While being an important part of the process

of treatment, IDC detection is an exceedingly time consuming and difﬁcult task.

In light of the above situation, the problem of classifying histopathological slides into

IDC(+) and IDC(-) has been delegated to ML using image processing techniques. ML relies

on the extraction of feature descriptors from the image followed by classiﬁcation. These

descriptors include color descriptors, like RGB histograms; statistical descriptors, like

Haralick features; and texture descriptors, like Speeded-Up Robust Features (SURF), etc.

Though signiﬁcant research has been devoted to classiﬁcation using either of the

feature descriptors, very little has been done to agglomerate the information furnished

by each descriptor. The alternative to ML is the usage of neural networks in DL, which

dispenses with the need for hand-crafted features. However, this comes at the cost of

greater complexity and the need for higher computational power. In addition, to train

a DL model properly, there is a requirement of a huge amount of input samples, which

sometimes may not be available for real-life problems.

Keeping the above facts in mind, in this paper, we attempted to classify the histopatho-

logical images into IDC(+) and IDC(-) using an ensemble of texture and statistical features.

The accuracy of texture features, such as Scale Invariant Feature Transform (SIFT) [

SURF [

], Oriented FAST and rotated BRIEF (ORB) [

] and Gray Level Co-Occurrence

Matrix (GLCM), as well as statistical features, such as Haralick features [

] are computed

using various standard ML algorithms, such as Random Forest (RF) [

], CatBoost (CB) [

]

and AdaBoost (AB) [

]. Such scores are then compared with the accuracy metrics for

various combinations of these features. The best results are obtained by ensembling all

features using a stacking method. However, as explained later, the stacking ensemble

has been tweaked to weed out redundant features using Pearson’s Correlation Coefﬁcient

based feature selection.

Keeping the above facts in mind, in this work, we made an exhaustive set of experi-

ments as well as comparisons using the well-known statistical and texture descriptors to

identify the best subset of features, which contribute to differentiate between cancerous and

non-cancerous cells. We also identiﬁed a set of ML classiﬁers that learn complimentary fea-

tures, which were aggregated to provide a better feature set. While an ensemble approach is

well-known in the literature; here, we have been able to ﬁnd the correct blend of statistical

and texture descriptors to achieve better results, and we have been able to even reduce the

dimension of the feature vector without compromising the classiﬁcation accuracy.

In a nutshell, the contributions of our proposed work are as follows:

An ensemble method of texture and statistical descriptors followed by classiﬁcation

using ML classiﬁers, like RF and CB, is proposed. The approach used for ensembling

is stacking.

The conventional method of stacking is modiﬁed to weed out potentially redundant

features using Pearson’s Correlation Coefﬁcient based feature selection—the details

of which are elaborated in Figure 1.

A thorough comparison enumerating the relative performances of SURF, SIFT, ORB,

GLCM and Haralick and various combinations of the same is presented.

A light-weight ML model is created, which is computationally inexpensive and

algorithmically easy to design. It is equally, if not more, accurate than other exist-

ing methods.

Sensors 2021, 21, 3628 4 of 17

(c)

Figure 1.

(

) The classiﬁer is trained on the training set. The trained classiﬁer is then used to predict

the outcome of the validation set as well as the test set. (

) The second stage wherein the validation

predictions for different classiﬁers are stacked to generate the new training features while the test

predictions for different classiﬁers are stacked to generate the new test features. Using these new

training and test features, the classiﬁer is evaluated. (

) Our model wherein the second stage is

modiﬁed and we perform feature selection on the new training and test features before model training

and testing.

The remaining paper is arranged as follows: Section 2 details the research endeavors

pertaining to the IDC(+) predictions found in the recent literature. Section 3 provides our

proposed model along with the relevant dataset description, while the experimental results

are enumerated and discussed in Section 4. Finally, the paper concludes in Section 5 along

with some future research directions.

2. Literature Survey

Over the last few years, researchers have suggested various methods for breast cancer

diagnosis in histopathology images. New innovative methodologies are being proposed

that include traditional machine learning and deep learning architectures. Many research

works have placed emphasis on feature extraction. Textural and statistical features are often

used. Doyle et al. [

] made use of 64 Gabor ﬁlter features, 15 statistical gray-level features,

graph-based features, 16 second-order statistical features and 24 nuclear features per each

of three color channels (hue, saturation and intensity) the in HSI color space utilizing three

distinct window sizes.

Dundar et al. [

] created an automatic classiﬁcation system for classifying breast

microscopic tissues into actionable subtypes (atypical ductal hyperplasia and ductal carci-

noma in situ) and usual ductal hyperplasia. Statistical features, like the perimeter, mean

gray-level intensity and the ratio of major to minor axis of the best-ﬁtting ellipse, were

utilized to model the cell size, shape and nucleoli appearance, respectively (histological

descriptors). Niwas et al. [

] proposed a method where they extracted ﬁrst-order statistical

and second-order statistical features.

With the help of the Log–Gabor complex wavelet bases, the color texture features of

the segmented nucleus were assessed. The aforementioned techniques include integrating

剩余16页未读，继续阅读

评论收藏

内容反馈

托比-马奎尔

粉丝: 1821

基于纹理和统计图像特征集成的计算机辅助乳腺癌检测

基于支持向量机的乳腺癌辅助诊断

乳腺癌细胞检测

乳腺肿瘤诊断数据集

resnet-使用resnet+SVM实现乳腺癌检测-算法.zip

基于机器学习的乳腺癌检测内含数据集.zip

基于纹理特征的图像检索

基于纹理特征的图像检索技术研究

基于纹理的图像检索源代码

基于码书和纹理特征的运动目标检测

基于纹理特征的图像检索系统设计

用KNN算法诊断乳腺癌

计算机辅助检测在乳腺癌X线检测中的价值

乳腺癌

基于Gabor滤波器的图像纹理特征提取

基于纹理的图像检索技术研究

基于纹理特征的图像检索.doc

基于纹理图像进行分类处理

基于深度学习的乳腺癌病理图像自动分类.pdf

基于纹理特征的图像分类识别

openCV纹理图像特征提取，比较两幅图像的相似度

基于纹理特征的图像检索方法

基于颜色和纹理特征的图像检索研究

基于纹理的图像检索源码matlab

基于matlab的图像纹理特征提取

【乳腺癌检测】医学乳腺癌检测处理系统（自适应中值滤波 GMM分类）【含Matlab源码 8770期】 (1).md

纹理特征提取的图像检索方法.

LBP.zip_纹理合成_纹理图像检测_纹理特征_纹理计数

STM32学习笔记

交换机旁挂ac

最新资源