Differential diagnosis of brucellar spondylitis and tuberculous spondylitis based on FS-T2WI sequence combined with machine learning
10.3760/cma.j.cn231583-20220628-00237
- VernacularTitle:基于FS-T2WI序列联合机器学习对布鲁氏菌性脊柱炎与结核性脊柱炎的鉴别诊断
- Author:
Tuxunjiang PAHATI
1
;
Laihong YANG
;
Xiong HE
;
Yushan CHANG
;
Wenya LIU
;
Yuwei XIA
;
Hui GUO
Author Information
1. 新疆医科大学第一附属医院影像中心，乌鲁木齐 830054
- Keywords:
Machine learning;
Brucellar spondylitis;
Tuberculous spondylitis;
Fat suppression;
Differential diagnosis
- From:
Chinese Journal of Endemiology
2023;42(5):356-362
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the performance of a predictive model based on fat suppression (FS)-T2WI sequence combined with machine learning in the differential diagnosis of brucellar spondylitis (BS) and tuberculous spondylitis (TS).Methods:The clinical and imaging data of 74 patients with BS and 81 patients with TS diagnosed clinically or pathologically in the First Affiliated Hospital of Xinjiang Medical University from January 2017 to January 2022 were retrospectively analyzed, and all patients underwent spinal magnetic resonance imaging (MRI) examination before treatment. Patients were randomly divided into a training group ( n = 123) and a testing group ( n = 32) in an 8 ∶ 2 allocation ratio, and radiomics feature extraction and dimensionality reduction analysis were performed on FS-T2WI sequence images. Four machine learning algorithms, including K-nearest neighbor (KNN), support vector machine (SVM), random forest (RF) and logistic regression (LR), were used to construct a radiomics model, and receiver operating characteristic (ROC) curve was used to analyze the differential diagnostic performance of each model for BS and TS. Results:A total of 1 409 radiomics features were extracted, and 7 related features were screened and included for identification of BS and TS, among which the Maximum2DDiameterColumn feature value showed a strong correlation, and there was a statistically significant difference between BS and TS patients ( P < 0.001). In the testing group, the area under the ROC curve (AUC) value of the SVM model for identifying BS and TS was 0.886, with a sensitivity of 0.53, a specificity of 0.88, and a diagnostic accuracy of 0.81; in the training group, the AUC value of the SVM model for identifying BS and TS was 0.811, the sensitivity was 0.68, the specificity was 0.72, and the diagnostic accuracy of the model was 0.78. Conclusion:The prediction model based on FS-T2WI sequence combined with machine learning can be used to identify BS and TS, and the diagnostic performance of SVM model is prominent and stable.
