To explore the value of texture features in magnetic resonance imaging (MRI) for predicting post-stroke cognitive impairment (PSCI) in elderly patients, and to analyze its correlation with the efficacy of PSCI drug therapy, providing a basis for selecting the timing and optimizing clinical drug intervention plans.

A retrospective analysis was conducted on the data of stroke patients admitted from January 2020 to January 2024. The data was divided into a training set (n=160) and a validation set (n=56) according to the data source. Collect PSCI standardized medication treatment data (including cholinesterase inhibitors, memantine, and combination therapy regimens) for all patients from baseline to cognitive assessment. After preprocessing of T1 weighted imaging (T1w) images, texture features were measured in the hippocampus and entorhinal cortex. Patients with high hippocampal kurtosis (≥ 2.5) and low hippocampal kurtosis were grouped and analyzed. A PSCI prediction and efficacy prediction model was constructed and validated based on drug treatment information. As a result, a total of 327 patients were included in this study, including 211 males and 116 females. The accuracy of the prediction model in the training set and validation set is 90% and 77%, respectively; Covariance analysis showed that there were differences (P<0.05) in hippocampal kurtosis, entorhinal cortex kurtosis, and entorhinal cortex deficit anomaly (IDM) between PSCI patients and non PSCI patients. The effective rate of donepezil treatment showed that the group with high hippocampal kurtosis was significantly higher than the group with low kurtosis (76.31% vs. 41.22%, P<0.01); Patients with IDM ≥ 0.4 in the entorhinal cortex showed better efficacy with combination therapy (donepezil+memantine) compared to monotherapy (82.13% vs. 58.28%, P<0.05). After optimizing the model and incorporating drug treatment factors, the validation set AUC increased to 0.82.

Conventional clinical MRI texture features are reliable early predictors of PSCI and are significantly associated with the efficacy of PSCI drug therapy. They can be combined with drug therapy information to construct more accurate prediction and efficacy evaluation models, providing imaging support for individualized drug intervention in PSCI.