AUTHOR=Yan Yuting , He Xiaodong , Xu Yuyun , Peng Jiaxuan , Zhao Fanfan , Shao Yuan 

TITLE=Comparison between morphometry and radiomics: detecting normal brain aging based on grey matter

JOURNAL=Frontiers in Aging Neuroscience

VOLUME=16

YEAR=2024

URL=https://www.frontiersin.org/journals/aging-neuroscience/articles/10.3389/fnagi.2024.1366780

DOI=10.3389/fnagi.2024.1366780

ISSN=1663-4365

ABSTRACT=<sec id="sec1"><title>Objective</title><p>Voxel-based morphometry (VBM), surface-based morphometry (SBM), and radiomics are widely used in the field of neuroimage analysis, while it is still unclear that the performance comparison between traditional morphometry and emerging radiomics methods in diagnosing brain aging. In this study, we aimed to develop a VBM-SBM model and a radiomics model for brain aging based on cognitively normal (CN) individuals and compare their performance to explore both methods’ strengths, weaknesses, and relationships.</p></sec><sec id="sec2"><title>Methods</title><p>967 CN participants were included in this study. Subjects were classified into the middle-aged group (<italic>n</italic> = 302) and the old-aged group (<italic>n</italic> = 665) according to the age of 66. The data of 360 subjects from the Alzheimer’s Disease Neuroimaging Initiative were used for training and internal test of the VBM-SBM and radiomics models, and the data of 607 subjects from the Australian Imaging, Biomarker and Lifestyle, the National Alzheimer’s Coordinating Center, and the Parkinson’s Progression Markers Initiative databases were used for the external tests. Logistics regression participated in the construction of both models. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were used to evaluate the two model performances. The DeLong test was used to compare the differences in AUCs between models. The Spearman correlation analysis was used to observe the correlations between age, VBM-SBM parameters, and radiomics features.</p></sec><sec id="sec3"><title>Results</title><p>The AUCs of the VBM-SBM model and radiomics model were 0.697 and 0.778 in the training set (<italic>p</italic> = 0.018), 0.640 and 0.789 in the internal test set (<italic>p</italic> = 0.007), 0.736 and 0.737 in the AIBL test set (<italic>p</italic> = 0.972), 0.746 and 0.838 in the NACC test set (<italic>p</italic> &lt; 0.001), and 0.701 and 0.830 in the PPMI test set (<italic>p</italic> = 0.036). Weak correlations were observed between VBM-SBM parameters and radiomics features (<italic>p</italic> &lt; 0.05).</p></sec><sec id="sec4"><title>Conclusion</title><p>The radiomics model achieved better performance than the VBM-SBM model. Radiomics provides a good option for researchers who prioritize performance and generalization, whereas VBM-SBM is more suitable for those who emphasize interpretability and clinical practice.</p></sec>