A Neuroradiomic Approach for Age-Group Classification of Healthy Human Brains

Neuroradiomics is an emerging subfield of radiomics that focuses on the extraction of high-dimensional quantitative measures from region of interests within the human neural tissue. This study aimed to extract radiomic features from the grey matter (GM) of structural brain magnetic resonance imaging (MRI) of seventy-five healthy control (HC) subjects, to classify them based on age-related differences. Subjects were classified as 'young' and 'old' using the median age as a reference, excluding those within a range of ten years around the median. T1-weighted anatomical scans were acquired for all participants, and MRI images were preprocessed before radiomic feature extraction. A total of 107 features, including shape, first order and texture, were collected. Several machine learning (ML) algorithms-Extra Tree Classifier (ETC), Random Forest (RF), XGBoost (XGB), Support Vector Machine (SVM), Linear Regression (LR), and K-Nearest Neighbor (KNN)-were employed to classify HCs in the two groups. Additionally, the SHapley Additive exPlanations (SHAP) analysis was carried out to identify the most informative radiomic features contributing to brain age estimation. Most classifiers achieved performance metrics exceeding 0.80, with LR yielding the best results: an accuracy of 0.86 (± 0.09), sensitivity of 0.88 (+0.12) and specificity of 0.84 (± 0.10), in terms of mean (±standard deviation). Finally, the SHAP analysis further revealed that texture features were predominant among the 15 most influential variables used in the training of models. In conclusion, the presented neuroradiomics approach shows promise for characterising age-related changes in the healthy brain using structural MRI.