Novel PET-driven brain mapping for diagnosis of Alzheimer’s disease: integrated feature extraction and ranking capabilities

Abstract

Early diagnosis of Alzheimer’s disease (AD) remains a significant challenge in modern healthcare. Positron emission tomography (PET) imaging combined with machine learning offers a powerful framework for this task, where brain atlases play a critical role in feature extraction. Although predefined atlases are widely used, they may provide suboptimal representations. This study introduces the novel PET-driven C-Atlas for AD diagnosis. Unlike conventional predefined brain atlases derived from healthy-control magnetic resonance imaging data, the C-Atlas is specifically designed for AD diagnosis and directly captures disease-relevant metabolic patterns from PET images. The C-Atlas learning consists of two phases: coefficient learning and atlas learning. In the first phase, PET images together with their corresponding labels (AD, MCI, NC) are used to train a linear support vector machine (SVM) classifier. The coefficients learned by SVM are subsequently employed in the second phase, where the simple linear iterative clustering algorithm is applied to learn the proposed atlas. Experimental results demonstrate that C-Atlas consistently outperforms predefined atlases. In addition, its construction process exhibits stable performance, and structural analysis reveals a moderate relationship between atlas organization and classification outcomes. Its superior and robust performance highlight its potential as a useful brain atlas and a promising tool for advancing early AD diagnosis and related neuroimaging applications.