Diffusion-Weighted Imaging-Based Radiomics Features and Machine Learning Method to Predict the 90-Day Prognosis in Patients With Acute Ischemic Stroke.

Abstract

Objectives: The evaluation of the prognosis of patients with acute ischemic stroke (AIS) is of great significance in clinical practice. We aim to evaluate the feasibility and effectiveness of diffusion-weighted imaging (DWI) image-based radiomics features and machine learning methods in predicting 90-day prognosis among patients with AIS.

Patients and methods: We enrolled a total of 171 patients with AIS in this study, including 134 patients with a good prognosis and 37 patients with a poor prognosis, and collected the patients' clinical and DWI image data. Radiomics features from manually sketched ischemic lesions were extracted using the Pyradiomics package of Python, and the best radiomics features were selected by a t test and the least absolute shrinkage and selection operator. The radiomics model and clinical model were constructed using support vector machine and logistic regression, respectively, and the predictive performance of each model was evaluated.

Results: We selected 9 features from a total of 851 radiomics features to build the final radiomics model. For predicting the poor prognosis of patients with AIS, the area under the curves, accuracy, sensitivity and specificity of the clinical model, radiomics model in the training set and radiomics model in the testing set were 0.865, 0.930 and 0.906, 81.3%, 92.0% and 90.0%, 81.1%, 76.0% and 75.0%, and 81.3%, 97.0% and 95.0%, respectively.

Conclusions: DWI image-based radiomics features and machine learning methods can accurately predict the 90-day prognosis of patients with AIS, and the radiomics model is superior to the clinical model in predicting prognosis.