A Multi-modal Data Platform for Diagnosis and Prediction of Alzheimer’s Disease Using Machine Learning Methods

Alzheimer’s an irreversible neurodegenerative disease with the most far-reaching impact, the most extensive, and the most difficult to cure in the world. It is also the most common disease of Alzheimer’s disease. With the rapid rise of data mining, machine learning and other fields, they have penetrated various disciplines. In particular, research in the field of AD is developing rapidly and has demonstrated strong vitality. In terms of data, Alzheimer’s Disease Neuroimaging Initiative (ADNI) researchers collect, verify and use a variety of data modalities as predictors of disease, including MRI and PET images, genetics, cognitive testing, cerebrospinal fluid and blood biomarkers, etc. Therefore, this paper uses a multi-task learning algorithm based on the ADNI data set to implement regression tasks and predict the cognitive scores of subjects in the next 3 years. This method can effectively assess the cognitive trends of patients in the future and aims to predict the progression of the disease. In addition, we used four different machine learning classification algorithms to conduct fusion research on AD multi-modal data, including MRI, PET, and cognitive scoring data. This method can determine the current patient’s cognitive stage, to achieve the effect of assisting doctors in diagnosis. Finally, we designed a multi-modal data platform technical architecture to standardize management and sharing of ADNI data and data obtained by offline medical institutions to improve the utilization and value of data. The design of the technical architecture proposed in this article is more easily scalable and compatible with other neurological diseases. Nowadays, the large amount of data being generated by AD can provide valuable solutions for the research of disease progression prediction and auxiliary diagnosis.