Metabolic networks on PET-based C-Atlas for diagnosis of Alzheimer's disease
9th International Conference on Multimedia Analysis and Pattern Recognition, MAPR 2026, Hue City, Vietnam, 13 - 14 Ağustos 2026, ss.472-477, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/mapr72750.2026.11685820
- Basıldığı Şehir: Hue City
- Basıldığı Ülke: Vietnam
- Sayfa Sayıları: ss.472-477
- Anahtar Kelimeler: Alzheimer's disease, brain atlas, FDG-PET, metabolic network, ROI selection, support vector machine
- Galatasaray Üniversitesi Adresli: Evet
Özet
Alzheimer's disease (AD) is the leading cause of dementia in elderly populations. Accurate and early diagnosis of AD has become an important objective of neuroimaging research. Among the available image modalities, 18F-fluorodeoxyglucose positron emission tomography (FDG-PET) is particularly well-suited to AD diagnosis. It has motivated the construction of metabolic networks that model AD as a disorder of interacting brain regions rather than isolated areas. The networks depend critically on the parcellation that defines their nodes. Yet, the atlases most commonly used in practice are derived from healthy anatomy or functional and do not reflect PET and AD specificity. In this paper, we propose a classification framework for AD detection that combines the PET-driven Coefficient-Atlas (C-Atlas) with kernel density estimate metabolic networks and a network-level region of interest (ROI) ranking. The proposed metabolic network pipeline achieves 93.1% accuracy for AD-vs-CN classification at M = 300 ROIs, surpassing the ROI-mean baseline. The proposed ROI ranking further boosts accuracy to 94.7% at only K* =77 regions. Within the same pipeline, C-Atlas achieves the highest classification accuracy among the four tested parcellations (AAL, Brainnetome, Schaefer, and C-Atlas) while requiring a more compact representation, confirming that C-Atlas provides a stronger basis for FDG-PET-based AD diagnosis.