Early Detection of Alzheimer’s Disease via Amyloid Aggregates: A Systematic Review of Plasma Spectral Biomarkers Using Machine Learning Approaches

Background: Alzheimer’s disease (AD) poses significant challenges in early diagnosisdue to the lack of reliable and non-invasive biomarkers. The aggregation ofβ-amyloid proteins and their fluorescence properties, alongside the Aβ42/Aβ40 ratioin cerebrospinal fluid (CSF) and blood, are promising indicators. Artificial intelligence(AI) has emerged as a powerful tool for analyzing these biomarkers, offeringimproved precision in detecting pathological changes and enabling earlier intervention.This systematic review aims to synthesize the existing literature on: Biomarkersfor the detection of β-amyloid aggregates; the relationship between β-amyloid aggregatelevels and the Aβ42/Aβ40 ratio in body fluids such as CSF and blood; the roleof machine learning techniques in the identification of β-amyloid aggregates and theanalysis of related data from experimental measurements.Methods: This systematic review will adhere to the Preferred Reporting Itemsfor Systematic Reviews and Meta-Analyses (PRISMA) guidelines. An exhaustivebibliographic search will be carried out in the following databases: Scopus, PubMed,Web of Science and IEEE Xplore. Articles will be screened based on predefinedinclusion and exclusion criteria. Data on biomarkers, aggregation levels, Aβ42/Aβ40ratios, and machine learning approaches will be extracted and synthesized.Results: 28 studies were selected for the review. From these works it was observedthat plasma Aβ42/Aβ40 showed consistent reductions (7%-18%) in PET-positive (PET+) individuals across studies, with mass spectrometry-based assaysyielding higher diagnostic accuracy (AUC 0.542–0.93, avg. 0.787) compared to immunoassaytechniques (AUC 0.540–0.877, avg. 0.708). CSF Aβ42/Aβ40 outperformedplasma measurements, demonstrating a larger differential (∼50% reductionin PET+ individuals). Fluorescence-based probes, particularly novel compoundssuch as AN-SP and CRANAD-28, showed superior sensitivity to amyloid oligomersand potential for in vivo imaging. AI/ML approaches, especially tree-based modelsand multichannel fluorescence sensor arrays, achieved high accuracy (> 90%),highlighting their potential for early Alzheimer’s detection.Conclusions: Plasma Aβ42/Aβ40 is a promising non-invasive biomarker fordetecting amyloid pathology, offering moderate to high diagnostic accuracy dependingon the assay used. However, CSF Aβ42/Aβ40 remains a superior predictor ofamyloid burden, with higher sensitivity, specificity, and stronger correlation withamyloid PET. While plasma biomarkers serve well as initial screening tools to reducethe dependence on costly PET scans, advanced fluorescence probes, particularlyratiometric and NIR dyes, demonstrate improved specificity and potential for in vivoimaging. AI-driven methodologies, including sensor arrays and non-invasive plasmaand IR-based techniques, show promise in enhancing early Alzheimer’s diagnostics.Combining multiple biomarkers with automated feature selection (e.g., AutoML,PCA, LDA) could significantly improve early detection strategies, emphasizing theneed for further validation to facilitate clinical adoption.