Development and validation of radiomics and deep transfer learning models to assess cognitive impairment in patients with cerebral small vessel disease

Abstract

Cognitive impairment in cerebral small vessel disease (CSVD) progresses subtly but carries significant clinical consequences, necessitating effective diagnostic tools. This study developed and validated predictive models for CSVD-related cognitive impairment using deep transfer learning (DTL) and radiomics features extracted from hippocampal 3D T1-weighted MRI. A total of 145 CSVD patients and 99 control subjects were enrolled in the study. We employed an automated algorithm to segment the hippocampus from 3D T1 images. Pre-trained deep learning networks were utilized to extract DTL features. Feature selection was performed using the Spearman rank correlation test and least absolute shrinkage and selection operator (LASSO) regression. Machine learning classification models, including Random Forest and Naive Bayes, were trained on the selected features. The predictive performance of these models was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) and decision curve analysis (DCA). The DTL model based on the ResNet101_32x8d network exhibited superior performance compared to other DTL models and the radiomics model, achieving an AUC of 0.847 (95 % CI: 0.691–1.000) and accuracy of 0.760. Furthermore, a combined model integrating ResNet101_32x8d and radiomic features further improved performance (AUC = 0.873, accuracy = 0.800), although the Delong test did not show statistical significance between models. These findings highlight that comprehensive data encompassing radiomics and DTL features showcase a robust predictive capability in distinguishing CSVD patients with cognitive impairment, offering insights for clinical applications despite limitations in sample size.