Weka-based classification and optimization of acute lymphocytic leukemia images
10.19745/j.1003-8868.2025022
- VernacularTitle:基于Weka的急性淋巴细胞白血病图像的分类与优化研究
- Author:
Xian-le SHI
1
;
Ting CHEN
1
;
Bao-lin HE
1
;
Yuan ZHOU
1
Author Information
1. 中国医学科学院血液病医院(中国医学科学院血液学研究所),实验血液学国家重点实验室,国家血液系统疾病临床医学研究中心,细胞生态海河实验室,天津 300020;天津医学健康研究院,天津 301600
- Publication Type:Journal Article
- Keywords:
Weka;
acute lymphoblastic leukemia;
machine learning;
image classification;
auxiliary diagnosis
- From:
Chinese Medical Equipment Journal
2025;46(2):10-15
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose a Weka-based method for classifying acute lymphoblastic leukemia(ALL)images,aiming to improve ALL cell classification accuracy and stability.Methods Firstly,totally 180 images were randomly selected from ALL-IDB2 subset of Acute Lymphoblastic Leukemia Image Database(ALL-IDB),including 90 images of patients and 90 images of healthy people;secondly,the image preprocessing was carried out using ImageJ software and image features were extracted such as texture,edge and shape;thirdly,image classification was implemented with four classifiers of Weka,including random forest(RF),Bayesian network(BN),J48 decision tree and sequential minimal optimization(SMO),and the key parameters of each classifier were optimized;finally,the performance of the classifiers was verified using 80 independent test images.Results Before parameter optimization,the accuracy of RF,J48 decision tree,BN and SMO classifiers was 94.3％,86.2％,83.6％and 83.0％,respectively.After optimization,the accuracy increased to 95.2％,86.3％,86.3％and 89.7％,respectively.After optimization,RF behaved the best on the independent test set with a classification accuracy of 90.0％,followed by SMO(81.3％),BN(81.3％)and J48 decision tree(75.0％).Conclusion The Weka-based ALL image classification method with a high accuracy is efficient and reliable for automated classification of ALL cell.[Chinese Medical Equipment Journal,2025,46(2):10-15]
