Related Experiment Video
Updated: May 10, 2026
Assessing Cortical Cerebral Microinfarcts on High Resolution MR Images
Published on: November 20, 2015
Hippocampal T1WI radiomics- and clinical feature-based models for predicting early mild cognitive impairment in
Xiaofeng Wang1, Ziao Xu1, Bohang Liu1
1Department of Neurosurgery, First Affiliated Hospital of Anhui Medical University, Hefei, China.
Introduction:
Mild cognitive impairment (MCI) represents the initial stage of dementia, and early diagnosis is crucial in clinical practice. This study aimed to investigate the predictive performance of three models based on clinical features, radiomics features of hippocampal T1-weighted imaging, and a combination of these features for identifying MCI in patients with secondary hydrocephalus.
Methods:
Of the 378 patients with secondary hydrocephalus, 124 were ultimately included in the study and divided into two cohorts: those with Mild Cognitive Impairment (MCI, n = 49) and those without MCI (n = 75). The samples were randomly stratified into a training set (34 MCI and 52 non-MCI patients) and a validation set (15 MCI and 23 non-MCI patients). Radiomic features from the bilateral hippocampi were extracted based on the region of interest, and the optimal parameters were selected through dimensionality reduction. Predictive models were constructed using clinical data, radiomic data, and a combination of both, with the radiomic score being utilized. The performance of each model was then assessed in both training and validation sets. Additionally, the diagnostic performance of the optimal model was compared with that of the Montreal Cognitive Assessment (MoCA) Scale.
Results:
In the clinical model, the disease course, serum uric acid, serum cystatin C, and the lateral ventricular temporal horn ratio emerged as independent risk factors for MCI following hydrocephalus. In the radiomics model, four optimal hippocampal features were identified. The AUC values for the clinical, radiomics, and combined models in the training/validation sets were 0.827 (0.736 ~ 0.919)/0.812 (0.666 ~ 0.957), 0.864 (0.790 ~ 0.937)/0.849 (0.724 ~ 0.974), and 0.937 (0.889 ~ 0.985)/0.907 (0.804 ~ 1.000), respectively. The combined model exhibited higher AUC values than the MoCA scale in both datasets. There was a significant difference in the training set, and while the validation set showed a consistent trend, it did not achieve statistical significance.
Conclusion:
The combined model achieved optimal performance and demonstrated superior predictive capabilities for MCI in the patients with secondary hydrocephalus outperforming other models.
