Prediction of Neurological Recovery in Patients with Primary Basal Ganglia Hemorrhage Based on Non-Contrast Computed Tomography Radiomics Combined with Clinical Features.

Abstract

This study aimed to predict neurological recovery in patients with basal ganglia spontaneous intracerebral hemorrhage (SICH) by integrating non-contrast computed tomography radiomics with clinical features. We retrospectively collected and analyzed clinical and imaging data of patients with primary basal ganglia hemorrhage. Radiomic features of hematoma and perihematomal edema were extracted to build a model for modified Rankin scale score changes. All cases were randomly divided into training and testing sets according to 8:2, the optimal hyperparameters were selected by 5-fold cross-validation for the training set to build the model. The radiomics model was constructed by Z-score homogenization, principal component analysis dimensionality reduction, relief for feature screening, and AdaBoost as a classifier. Among 162 patients, age (P = 0.017) and hypertension history (P = 0.030) independently indicated poor outcome. The area under the curve values of the radiomics model in the training and testing datasets were 0.873 (95% confidence intervals [CIs]: 0.8121-0.9281) and 0.8549 (95% CI: 0.7085-0.9648), respectively. The area under the curve of the combined model constructed by clinical and radiomics features was slightly lower than that of the radiomics feature model of 0.8632 (95% CI: 0.8005-0.9121) and lower than that of the radiomics feature model of 0.8633 (95% CI: 0.8006-0.9122). The decision curve analysis curve demonstrated clinical benefit of both models. Based on non-contrast computed tomography radiomics and clinical characteristics, a model can be constructed into a quantitative clinical assessment tool to predict the status of neurological recovery in patients with SICH.