Prediction models for cognitive impairment in middle-aged patients with cerebral small vessel disease

This study aims to develop hippocampal texture model for predicting cognitive impairment in middle-aged patients with cerebral small vessel disease (CSVD).

The dataset included 145 CSVD patients (Age, 52.662 ± 5.151) and 99 control subjects (Age, 52.576±4.885). An Unet-based deep learning neural network model was developed to automate the segmentation of the hippocampus. Features were extracted for each subject, and the least absolute shrinkage and selection operator (LASSO) method was used to select radiomic features. This study also included the extraction of total intracranial volume, gray matter, white matter, cerebrospinal fluid, white matter hypertensit, and hippocampus volume. The performance of the models was assessed using the areas under the receiver operating characteristic curves (AUCs). Additionally, decision curve analysis (DCA) was conducted to justify the clinical relevance of the study, and the DeLong test was utilized to compare the areas under two correlated receiver operating characteristic (ROC) curves.

Nine texture features of the hippocampus were selected to construct radiomics model. The AUC values of the brain volume, radiomics, and combined models in the test set were 0.593, 0.843, and 0.817, respectively. The combination model of imaging markers and hippocampal texture did not yield improved a better diagnosis compared to the individual model (p > 0.05).

The hippocampal texture model is a surrogate imaging marker for predicting cognitive impairment in middle-aged CSVD patients.