Alzheimer's disease diagnosis from single and multimodal data using machine and deep learning models: Achievements and future directions

Alzheimer's Disease (AD) is the most prevalent and rapidly progressing neurodegenerative disorder among the elderly and is a leading cause of dementia. AD results in significant suffering for patients, impairing their ability to perform simple daily tasks and ultimately leading to death. Despite being identified over 100 years ago, AD is still not well understood, and considerable research efforts are ongoing to achieve early diagnosis, effective treatment, and prognosis. Unfortunately, no effective drug currently exists to cure or reverse the disease. Instead, current efforts focus on the early prediction of disease onset to provide timely treatments and delay its progression, particularly during the prodromal stage known as mild cognitive impairment (MCI). This requires measurable indicators (biomarkers) to assess subjects through brain imaging, cerebrospinal fluid analysis, genetic analysis, and/or blood testing. Recent advances in machine learning and deep learning techniques have significantly improved the efficiency of computer diagnostic tools in identifying disease-related biomarkers from single or multimodal data. In this paper, we first present the main framework for building disease diagnosis models, providing a detailed description of each step, followed by an overview of the different machine learning and deep learning models used in each phase. Next, we discuss the various criteria for constructing AD diagnosis models and explore various multimodal data fusion techniques. We then review the latest studies employing both machine learning and deep learning methods with single-modal or multi-modal biomarkers. Finally, we discuss existing challenges, potential solutions, and future research directions to further enhance the diagnostic performance of AD.