Deep learning techniques for automated dementia diagnosis using neuroimaging modalities: a systematic review

Abstract

Dementia is a condition that often comes with aging and affects how people think, remember, and behave. Diagnosing dementia early is important because it can greatly improve patients' lives. This systematic review looks at how deep learning (DL) techniques have been used to diagnose dementia automatically from 2012 to 2023. We explore how different DL methods like Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Deep Neural Networks (DNN) are used to diagnose types of dementia such as Alzheimer's, vascular dementia, and Lewy body dementia. We also discuss the difficulties of using DL for diagnosing dementia, like the lack of large and varied datasets and the challenge of applying models to different groups of people. These issues indicate the need for more dependable and understandable models that consider a wide range of patient characteristics and biomarkers. Longitudinal studies are also needed to understand how the disease progresses and how treatments work. Collaboration among researchers, doctors, and data scientists is crucial to ensure DL models are scientifically sound and effective in clinical settings. In summary, DL techniques show promise for automated dementia diagnosis and could improve how accurately and efficiently it is diagnosed in practice. However, further research is needed to address the challenges highlighted in this review.