Early diagnosis of Alzheimer’s disease using machine learning and blood biomarkers

Alzheimer’s disease imposes a substantial socioeconomic burden on families and society, while diagnosis at the Mild Cognitive Impairment stage remains critical for effective prevention and management. Current machine learning approaches, however, are predominantly reliant on invasive neuroimaging or cerebrospinal fluid analyses, thereby facing inherent diagnostic limitations. Blood biomarkers provide a minimally invasive alternative, yet their integration with explainable ML frameworks for early detection remains underexplored, particularly in the context of scalable clinical implementation. This study aimed to develop a classification model using blood biomarkers and machine learning to diagnose mild cognitive impairment as an early stage of Alzheimer’s disease. Blood biomarker data from 119 healthy controls and 672 MCI patients were collected from the Alzheimer’s Disease Neuroimaging Initiative database. Feature selection was performed using correlation-based and wrapper methods. Seven machine learning algorithms were used to build classification models and compare their performance, followed by constructing a stacking model. SHAP was used for model interpretation. The stacking model, combining Adaboost, Xgboost, and random forest with a support vector machine as the meta-model, achieved a sensitivity of 0.94 and an accuracy of 0.93. AgRP had the most significant impact on classification, followed by Eotaxin-3. This study presents an interpretable machine learning model that precisely identifies MCI using blood biomarkers. By not only demonstrating high diagnostic accuracy but also elucidating the roles of key biomarkers, this work establishes the significant potential of our approach as a scalable and minimally invasive tool for early clinical screening.