Artificial Intelligence-Assisted Volumetric Brain Analysis Correlated with CSF Biomarkers in Alzheimer's Disease: A
Pukovisa Prawiroharjo1,2,3, Amelia Nur Vidyanti4, Yuliarni Syafrita5
1Department of Neurology, Faculty of Medicine, Universitas Indonesia, Jakarta 10430, Indonesia.
None:
Background/Objectives: Alzheimer's disease (AD) is a leading cause of dementia globally, yet standard diagnostic markers like cerebrospinal fluid (CSF) analysis and molecular imaging are invasive and resource-intensive. While artificial intelligence (AI)-based volumetric magnetic resonance imaging (MRI) offers a scalable and non-invasive alternative, data correlating these structural metrics with fluid biomarkers and cognitive status in Southeast Asian populations are scarce. This study addresses this critical gap by examining the within-cohort relationship between CSF biomarkers and regional brain volumes derived from AI-assisted MRI in Indonesian patients with clinically diagnosed AD, providing novel data for an underrepresented population. Methods: Twenty-one AD patients from three national referral hospitals in Indonesia underwent lumbar puncture for CSF biomarker analysis and 3 Tesla structural brain MRI. Brain volumes were analyzed using United Imaging Intelligence software, focusing on AD-relevant regions (hippocampus, entorhinal cortex, parahippocampus, precuneus, and posterior cingulate cortex [PCC]). Results: Spearman's correlation revealed significant positive associations between CSF Aβ42 levels and several brain regions. Strong correlations were found with the right entorhinal volume indexed to intracranial volume (VICV) (r = 0.601, p = 0.004), right PCC VICV (r = 0.603, p = 0.004), right entorhinal volume (r = 0.533, p = 0.013), and right hippocampus VICV (r = 0.503, p = 0.020). Furthermore, MoCA-InA scores demonstrated highly significant positive correlations with CSF Aβ42 concentrations (r = 0.720, p < 0.001), right Hippocampus VICV (r = 0.703, p < 0.001), and right PCC VICV (r = 0.695, p < 0.001). No significant correlations were found between CSF pTau or the pTau/Aβ42 ratio and regional volumes. Conclusions: These results highlight the entorhinal cortex and PCC as early affected regions where CSF Aβ42 correlates with preserved volume, supporting their role as structural markers in early AD. The absence of pTau associations may reflect early-stage pathology or limitations of cross-sectional volumetry. In resource-limited settings, AI-assisted volumetric MRI demonstrates potential utility as a non-invasive tool for stratifying amyloid-associated brain atrophy and staging disease severity.
