Related Experiment Video
Updated: May 12, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Machine learning based on radiomics for discriminating sellar region langerhans cell histiocytosis from germ cell
Hongting Jiang1, Zanyong Tong1, Yu Luo1
1Department of Radiology, Children's Hospital of Chongqing Medical University, National Clinical Research Center for Children and Adolescents' Health and Diseases, Ministry of Education Key Laboratory of Child Development and Disorders, International Science and Technology Cooperation Base of Child Development and Critical Disorders, Chongqing Key Laboratory of Child Neurodevelopment and Cognitive Disorders, Chongqing, China.
Background:
Differentiating sellar region germ cell tumors (GCTs) from Langerhans cell histiocytosis (LCH) is challenging due to highly similar MRI features, especially in tumor marker-negative patients. In this study, we aimed to develop and validate a radiomics model to distinguish tumor marker-negative sellar GCTs from LCH.
Methods:
This retrospective study enrolled a total of 93 patients diagnosed pathologically or by therapeutic diagnosis at our single institution between April 2012 and April 2024, including 40 cases of LCH and 53 cases of GCTs. Radiomics features extracted from multiparametric MRI, including T1-weighted imaging (T1WI) and T2-weighted imaging (T2WI). We manually segmented the regions of interests (ROIs) of tumors. Feature selection was subsequently performed using LASSO regression with five-fold cross-validation. We have chosen three machine learning classifiers-Support Vector Machine (SVM), Random Forest (RF), and Logistic Regression (LR) to construct models based on the 7 features which were retained. Additionally, by integrating clinically significant features and imaging semantic features, classification models based on radiomics features, imaging semantic features, and clinical features were developed separately. There are 7 models in total. Furthermore, combined prediction models were constructed based on different fusion feature sets, respectively. The performance of the diagnostic model was evaluated using the receiver operating characteristic (ROC) curve. The mean area under the curve (AUC), sensitivity, specificity, accuracy, and F1-score were calculated for both the development set and the test set. Differences in AUC between models were assessed using DeLong's test, and the resulting P-values were adjusted using the Bonferroni false discovery rate (FDR) correction method. Code available upon request.
Results:
The best diagnostic performance was achieved by the combined model of radiomics with clinical features and imaging semantic features using the RF classifier, with an AUC value of 0.81. A statistically significant difference (p < 0.05) was confirmed by the DeLong test, indicating robust diagnostic capability.
Conclusion:
Radiomics-based machine learning is a promising, non-invasive approach to distinguish tumor marker-negative sellar GCTs from LCH, which has good predictive performance and may help with treatment decision-making.