The Role of Multimodal Machine Learning in Alzheimer’s Disease: A Review

Background and aims: Alzheimer’s disease is a progressive brain disorder that affects millionsof people worldwide. It is the most common cause of dementia and is characterized by a declinein cognitive function, memory loss, and behavioral changes. Despite extensive research efforts,Alzheimer’s disease remains incurable, making early detection and diagnosis crucial for effectivetreatment and management of the disease. Machine learning models have shown great potential inthe accurate diagnosis and prediction of brain disorders, including Alzheimer’s disease. However,the accuracy of these models can be further improved by using multimodal data, which combinesmultiple types of data such as medical images, clinical assessments, and genetic information.This review aims to summarize the state-of-the-art research on Alzheimer’s disease that utilizesmachine learning techniques with multimodal medical data.Method: The following search strategy has been used in PubMed, Web Of Science(WOS), andIEEE Xplore databases:((neuroimaging) OR (CT scan) OR (PET scan) OR (MRI) OR (radiology)OR (Electroencephalogram)) AND (multimodal*) AND ((“deep learning”) OR (“Computervision”) OR (“Neural Network*”) OR (“machine learning”) OR (“artificial intelligence”)) AND(Alzheimer’s disease). To conduct our literature review, we included all the words in the title/abstractas part of our search strategy and searched PubMed using relevant MeSH terms. The studyexcluded duplicated and non-journal articles. Title and abstract screening resulted in the exclusionof review articles and irrelevant publications. During the full-text screening, duplicated articles,irrelevant publications, and those lacking full-text availability were also excluded.Results: Different studies utilized different combinations of data modalities such as structuralMRI (sMRI), resting-state functional MRI (rs-fMRI), different types of Positron Emission Tomography(PET) such as amyloid-PET, genetic markers such as single-nucleotide polymorphisms(SNPs), cognitive scores, and CSF biomarkers. Additionally, different machine learning modelssuch as support vector machines(SVMs), convolutional neural networks(CNNs), recurrentneural networks(RNNs), and random forests have been deployed. Studies comparing the outcomesof utilizing multimodal data and unimodal data have demonstrated superior performancewhen utilizing multimodal data. Additionally, using the Alzheimer’s Disease Neuroimaging Initiative(ADNI) database in most studies under review facilitated direct comparisons of differentmodels’ performance, aiding in identifying the optimal approach.Conclusion: This review highlights the important role of multimodal machine learning in theearly detection and diagnosis of Alzheimer’s disease. An evaluation was conducted on the performanceof the models recommended in the reviewed literature. Additionally, the assessmentencompassed the analysis of diverse data modalities utilized, the type of machine learning modelimplemented, and the evaluation of the utilized datasets. In conclusion, the application of multimodalmachine learning presents a promising opportunity for revolutionizing the timely detectionand diagnosis of Alzheimer’s disease, signifying a hopeful avenue for future research in this field.

کلیدواژه ها:

Alzheimer’s disease ، multimodal ، machine learning ، neuroimaging