Machine Learning-Based Perihematomal Tissue Features to Predict Clinical Outcome after Spontaneous Intracerebral Hemorrhage

To explore whether radiomic features of perihematomal tissue can improve the forecasting accuracy for the prognosis of patients with an intracerebral hemorrhage (ICH). In total, 118 ICH patients were retrospectively studied that had a clinical and radiological diagnosis of spontaneous ICH. The functional outcome 3 months after ictus was measured using the modified Rankin Scale (mRS), which was divided into good (mRS ≤ 2) and poor outcomes (mRS > 2). A total of 2260 radiomics features were obtained from non-contrast computer tomography (NCCT) images, with 1130 features extracted from the hematoma and the hematoma plus perihematoma. The high-dimensional data was modeled by a logistic regression algorithm and the accuracy of the model was verified by five-fold cross-validation. The predictive performance of radiomics models was assessed by the area under the receiver operating characteristic (ROC) curve. In the test set, the mean ROC area under the curve (AUC) of the hematoma set to predict the prognosis of ICH was 0.83, and the specificity and sensitivity were 78% and 81%, respectively. When the hematoma and perihematomal tissue were combined, the mean AUC increased to 0.88, and the specificity and sensitivity reached 85% and 84%, respectively. The hematoma plus perihematoma model showed a significantly higher AUC and specificity. Analysis of the hematoma and perihematomal tissue NCCT-based radiomics could potentially identify the progression of a hematoma more accurately and could be a valuable clinical target to enhance the prediction of outcomes in patients with ICH.