Deep learning-based approaches to the Evaluation of Alzheimer's Neuropathologies in fMRI Data

Deep learning-based approaches to the Evaluation of Alzheimer's Neuropathologies in fMRI Data
Mohammadreza Elhaie,^1,* Abolfazl Koozari,² Mahmood Mohammadi Sadr,³ Iraj Abedi,⁴ Daryoush Shahbazi-Gahrouei,⁵ Ahmad Shanei,⁶
1. Department of Medical Physics, School of Medicine Isfahan University of Medical Sciences
2. Department of Medical Physics, School of Medicine Ahvaz Jundishapur University of Medical Sciences
3. Department of Medical Physics, School of Medicine Isfahan University of Medical Sciences
4. Department of Medical Physics, School of Medicine Isfahan University of Medical Sciences
5. Department of Medical Physics, School of Medicine Isfahan University of Medical Sciences
6. Department of Medical Physics, School of Medicine Isfahan University of Medical Sciences
Introduction: Alzheimer’s disease (AD) is the most common neurodegenerative disease. Several issues correlate with AD, such as a lack or loss of memory, disorientation, and loss of time comprehension. Recent research suggests that the susceptibility of the human population to AD has increased, and early prediction of AD can help control the symptoms drastically. Functional Magnetic Resonance Imaging (fMRI) is a neuroimaging technique that can distinguish between active and non-active brain regions based on blood flow, also known as the “Blood Oxygen Level Dependent” or “BOLD MRI”. On the other hand, the increased use of artificial intelligence, such as deep learning or machine learning algorithms, in conjunction with imaging in the healthcare field has been established significantly. The automation in image analysis such as “classification” has reached its acceptability for early diagnosis in medical imaging. The main objective of this review is to provide the reader with a clear picture of the role of deep-learning algorithms in the prediction of Alzheimer's disease based on fMRI data.
Methods: We searched for the following keywords in the Google Scholar database: “Functional Magnetic Resonance” OR “fMRI” “Deep learning” AND “Alzheimer's disease”. We also limited the publication dates of the articles to after 2023 to complete the analysis of the latest literature. The search strategy produced 3170 articles. We excluded irrelevant articles based on title and abstract screening. We included 51 articles reviewed in this study.
Results: We investigated the following parameters: accuracy, classification method, data source, and number of patients included in the articles. Most studies showed a significant percentage of accuracy (84%). The number of patients in this study was between 60-389 patients. Most studies acquired their data from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Almost all studies preprocessed fMRI data before modeling.
Conclusion: In this study, we investigated the correlation between deep learning and fMRI for disease detection. Most studies applied various DNN methods to detect AD at different stages. We reported that most studies showed high accuracy (84%) for AD detection. In addition, most studies have shown that deep learning methods are superior to machine-learning methods. We conclude that deep learning algorithms are important for initial-stage Alzheimer's disease detection.
Keywords: Deep learning, Alzheimer's disease,Functional Magnetic Resonance Imaging