Reliability of radiomics features as imaging biomarkers for evaluating brain aging: A study based on myelin protein and diffusion tensor imaging

Abstract

Radiomics has made considerable progress in neurodegenerative diseases. However, previous studies only explored the feasibility of radiomics in clinical applications. Therefore, the objective of this study was to obtain the most relevant radiomics features with the aging changes of myelin proteins and compare their diagnostic performances with the diffusion tensor imaging (DTI) parameters to identify the reliability of these features as imaging biomarkers for assessing brain aging. Thirty middle-aged and thirty old-aged mice were assigned to the training set to explore the most relevant features of myelin proteins and their diagnostic performances. Ten middle-aged and ten old-aged mice were assigned to the testing set to further validate the reproducibility of the features. T2-weighted imaging and DTI were conducted to obtain white matter radiomics features and DTI parameters. Myelin proteins, including proteolipid protein (PLP) and myelin basic protein (MBP), were examined by immunofluorescence staining in the regions of the whole brain, cortex, corpus callosum, striatum, and anterior commissure. The Pearson correlation analysis was used to observe the correlations between radiomics features and myelin proteins. The four most relevant features with the top four correlation coefficients were selected to compare their diagnostic performances with the DTI parameters, including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AxD), and radial diffusivity (RD). Wavelet-HLL_glszm_ZoneEntropy, wavelet-HLL_gldm_DependenceEntropy, wavelet-LHL_glszm_ZoneEntropy, and log-sigma-2-0-mm-3D_gldm_DependenceEntropy were the four most relevant features, which had moderately significant correlations with PLP. The area under the receiver operating characteristic curves (AUC) of the four features were 0.940, 0.917, 0.831, and 0.964 in the training set, and 0.880, 0.840, 0.860, and 0.880 in the testing set. The AUCs of FA, MD, AxD, and RD were 0.864, 0.743, 0.673, and 0.778 in the training set, and 0.780, 0.710, 0.670, and 0.730 in the testing set. These results demonstrated that radiomics features of white matter displayed significant correlations with myelin proteins and their performances were comparable or even superior to DTI parameters, which ensured their reliability as non-invasive imaging biomarkers for evaluating brain aging.