AUTHOR=Yang Qingning , Sun Jun , Guo Yi , Zeng Ping , Jin Ke , Huang Chencui , Xu Jingxu , Hou Liran , Li Chuanming , Feng Junbang 

TITLE=Radiomics Features on Computed Tomography Combined With Clinical-Radiological Factors Predicting Progressive Hemorrhage of Cerebral Contusion

JOURNAL=Frontiers in Neurology

VOLUME=Volume 13 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2022.839784

DOI=10.3389/fneur.2022.839784

ISSN=1664-2295

ABSTRACT=Background: Traumatic brain injury (TBI) is the main cause of death and severe disability in young adults worldwide. Progressive haemorrhage (PH) worsens the disease and can cause a poor neurological prognosis. Radiomics analysis has been used for haematoma expansion of hypertensive intracerebral haemorrhage. This study attempts to develop an optimal radiomics model based on noncontrast CT to predict PH by machine learning (ML) methods and comparing its prediction performance with clinical-radiologic models.
Methods: We retrospectively analysed 199 TBI patients, including 98 PHs and 101 non-PHs, whose data were randomized into a training set and a test set at a ratio of 7:3. 10 different machine learning methods were used to predict PH. Univariate and multivariate logistic regression were implemented to screen clinical-radiologic factors and to establish a clinical-radiologic model. Then, a combined model combining clinical-radiologic factors with the radiomics score was constructed. The ROC-AUC, accuracy and F1 score, sensitivity and specificity were used to evaluate the models.
Results: Among the 10 various ML algorithms, the support vector machine (SVM) had the best prediction performance based on 14 radiomics features, including the AUC (training set 0.878, test set 0.849) and accuracy (training set 0.820, test set 0.792). Among the clinical and radiologic factors, the onset-to-baseline CT time, the scalp haematoma and fibrinogen were associated with PH. The radiomics model’s prediction performance was better than the clinical-radiologic model, while the predictive nomogram combining the radiomics features with clinical-radiologic characteristics performed best.
Conclusions: The radiomics model outperformed the traditional clinical-radiologic model in predicting PH. The nomogram model of the combined radiomic features and clinical-radiologic factors is helpful tool for PH.Introduction