104 - Imaging features to predict the onset time of acute stroke: Model construction and efficacy evaluation research.

To explore the value of a machine learning model constructed based on DWI image features to predict the onset time of acute stroke. A retrospective analysis included 317 acute stroke patients treated in the Central Hospital of Shaoyang from January 2022 to January 2023. All patients were completely randomized into training (221) and internal testing (96) based on a 7:3 ratio; patients were divided into 4.5h and> 4.5h according to the onset time. The high signal area of the acute infarction on the DWI image was outlined by the chief physician.107 image features were extracted by 3D SLICER software, using F test correlation analysis and lasso regression, and the prediction model was constructed using a random forest classifier and validated in the test group. The efficacy of the prediction model was evaluated using the receiver operating characteristic curve and the area under the curve (Area under the curve, AUC). To compare the area under the ROC curve (AUC) and accuracy of manual identification and machine learning models to predict the onset time of acute stroke. ROC analysis showed that the AUC of manual identification predicting the onset of acute stroke was 0.554 (CI:0.44-0.67) and the AUC of machine learning model was 0.848(CI:0.74-0.95). DWI-based machine learning model predicts the onset time of acute stroke is significantly better than manual identification. In clinical practice, the prediction can be completed by imaging technicians before the end of MRI scan to provide more information for the diagnosis and treatment of stroke patients with unknown onset time. [ABSTRACT FROM AUTHOR]