Use of radiomic features and support vector machine to discriminate subjective cognitive decline and healthy controls

Subjective cognitive decline (SCD) is a high-risk preclinical stage in the progress of Alzheimer's disease (AD). Its timely diagnosis is of great significance for older adults. Though multi-parameter magnetic resonance imaging (MPMRI) is a noninvasive neuroimaging technique to detect SCD, the lack of biomarkers and computed aided diagnosis (CAD) tools is a major concern for its application. Radiomics, a high-dimensional imaging feature extraction method, has been widely used for identifying biomarkers and developing CAD tools in oncological studies. Therefore, in this study, we aimed to investigate whether the radiomic approach could be used for the diagnosis of SCD. In the proposed radiomic approach, we mainly performed four steps: image preprocessing, feature extraction and screening, and classification. The dataset from Xuanwu Hospital, Beijing, China, was used in this study, including 105 healthy controls (HC) and 130 SCD subjects. All subjects were divided into one training & validation group and one test group. We extracted 30128 radiomic features from MPMRI of each subject. The t-test, autocorrelation, and Fisher score were performed for feature selection, and we deployed the support vector machine (SVM) for classification. The above process was performed 100 times with 5-fold cross-validation. The results showed that the accuracy, sensitivity, and specificity of classification was 89.03%+/- 5.37%, 85.44%+/- 9.28% and 91.97%+/- 6.38% in the validation set and 84.70%+/- 4.68%, 86.98%+/- 10.49% and 82.59%+/- 7.07% in the test set. In conclusion, this study has shown that the radiomic approach could be used to discriminate SCD and HC with high accuracy and sensitivity effectively.