Autor:innen:
K. Waschkies (Göttingen, DE)
J. Soch (DE)
M. Darna (DE)
A. Richter (DE)
E. Düzel (DE)
F. Jessen (DE)
J. Wiltfang (DE)
B. Schott (DE)
J. Kizilirmak (DE)
Background: During the last two decades, machine learning approaches have been used to test the diagnostic value of variables for Alzheimer’s dementia (AD) and increased AD risk, automated classification, and for prediction of conversion. While biomarkers from cerebrospinal fluid (CSF) are the best-established predictors, less invasive candidate predictors show considerable association with AD and increased AD risk. Importantly, less invasive markers are desirable to identify at-risk individuals, such as patients with mild cognitive impairment (MCI) or subjective cognitive decline (SCD), to improve early prevention or therapeutic intervention.
Methods: We comparatively evaluated the diagnostic value of self-report-based personality traits, anxiety, and depression scores, as well as resting-state fMRI as a non-invasive neurophysiological measure, and established CSF markers (Aß42/Aß40-ratio, pTau181, total Tau) in classifying healthy controls (HC) and patients with SCD, MCI, and mild AD, using multi-class support vector machine classification. Predictors were assembled into distinct feature sets and their classification performances were assessed. A total of 733 participants (189 HC, 338 SCD, 132 MCI, 74 AD) from the multi-center DZNE-Longitudinal Cognitive Impairment and Dementia Study (DELCODE) were included.
Results: We found that a combination of personality, anxiety, and depression scores descriptively performed best at classifying HC and MCI while also yielding equal overall decoding accuracy compared to CSF biomarkers. CSF biomarkers performed best in classifying SCD and AD. However, classification of SCD and MCI was relatively poor among all feature sets.
Summary and Conclusion: Our results suggest that SCD and MCI are heterogeneous groups, pointing out the importance of optimizing their diagnosis criteria. Moreover, CSF biomarkers, personality, depression, and anxiety indicate complementary value for class prediction, to be evaluated in future research.