Multicenter study on distinguishing long bone osteosarcoma from Ewing sarcoma based on CT image histogram and texture feature analysis
10.3760/cma.j.cn115455-20240311-00227
- VernacularTitle:基于CT图像直方图和纹理特征分析鉴别长骨骨肉瘤与尤文肉瘤的多中心研究
- Author:
Jianwei LI
1
;
Jingzhen HE
;
Jiuming JIANG
;
Sheng DING
;
Libin XU
;
Sijie HU
;
Chengyi JIANG
;
Li ZHANG
;
Meng LI
Author Information
1. 国家癌症中心/国家肿瘤临床医学研究中心/中国医学科学院北京协和医学院肿瘤医院影像诊断科，北京 100021
- Keywords:
Osteosarcoma;
Ewing sarcoma;
Tomography, X-ray computed;
Texture analysis
- From:
Chinese Journal of Postgraduates of Medicine
2024;47(10):875-880
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the application value of histogram and texture feature analysis based on CT images in distinguishing long bone osteosarcoma (OS) from Ewing sarcoma (ES).Methods:A retrospective collection of 25 patients with long bone osteosarcoma and 25 patients with Ewing sarcoma confirmed by surgery and pathology in National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Qilu Hospital of Shandong University and Nanjing Drum Tower Hospital, Nanjing University Medical School, from March 2018 to May 2023 was conducted. All patients were randomly divided into a training set (21 cases of OS and 19 cases of ES) and a validation set (4 cases of OS and 6 cases of ES) in an 8∶2 ratio. The region of interest (ROI) on CT images to extract texture feature parameters was manually sketched. Random forest and least absolute shrinkage and selection operator (LASSO) algorithm were used for feature screening. Logistic regression (LR), random forest (RF), support vector machine (SVM) and K-nearest neighbor (KNN) classifiers were used to establish models respectively. Receiver operating characteristic (ROC)curve was drawn and area under the curve (AUC) was calculated to evaluate the diagnostic efficiency of the four models.Results:A total of 100 texture parameters were extracted from CT images, and 8 feature parameters (maximum 3D diameter, 10th percentile, kurtosis, maximum pixel intensity value, inverse normalization, grayscale level variance, long range high grayscale emphasis, and low grayscale area emphasis) were obtained through screening. Four classifiers were used to establish models, and the AUC values of the four models (LR, RF, SVM, KNN) in the validation group were 0.92, 0.79, 0.83, and 0.73, respectively. LR and SVM classifier algorithm trains models had high diagnostic efficiency, with an accuracy of 90%, sensitivity of 83%, specificity of 100%, and AUC of 92% for the LR classifier validation set; the accuracy of SVM classifier validation set was 80%, sensitivity was 67%, specificity was 100%, and AUC was 83%.Conclusions:LR and SVM models have high value in distinguishing OS and ES.
