The differential diagnosis between mild cognitive impairment (MCI) and Alzheimer's disease (AD) has been highly demanded for its effectiveness in preventing and contributing to early diagnosis of AD. To this end, we developed a single plasmonic asymmetric nanobridge (PAN)-based biosensor to differentially diagnose MCI and AD by quantitative profiling of phosphorylated tau proteins (p-tau) in clinical plasma samples, which revealed a significant correlation with AD development and progression. The PAN was designed to have a conductive junction and asymmetric structure, which was unable to be synthesized by the traditional thermodynamical methods. For its unique morphological characteristics, PAN features high electromagnetic field enhancement, enabling the biosensor to achieve high sensitivity, with a limit of detection in the attomolar regime for quantitative analysis of p-tau. By introducing support vector machine (SVM)-based machine learning algorithm, the improved diagnostic system was achieved for prediction of healthy controls, MCI, and AD groups with an accuracy of 94.47 % by detecting various p-tau species levels in human plasma. Thus, our proposed PAN-based plasmonic biosensor has a powerful potential in clinical utility for predicting the onset of AD progression in the asymptomatic phase.