A quantitative analysis of Alzheimers Disease and construction of an early Alzheimers detection deep learning system (EADDLS) using MRI data via machine learning along with ADmod: spatiotemporal-aware brain-amyloidβ growth model, using deep encoder-decoder networks about MRI

Alzheimers disease (AD) presents a significant societal challenge, with no current cure and an increasing prevalence among older adults. This study addresses the pressing need for early detection by harnessing the potential of machine learning applied to longitudinal MRI data. The dataset, sourced from the Open Access Series of Imaging Studies (OASIS) project, comprises MRI records of 150 subjects aged 60 to 96, each scanned at least once. Notably, 72 subjects were classified as Nondemented, 64 as Demented, and 14 underwent a transition from Nondemented to Demented, forming the Converted category. What we propose is to develop a machine learning sound model capable of predicting the progression of mild cognitive impairment, leveraging key biomarkers extracted from MRI data. The chosen biomarkers include years of education (EDUC), socioeconomic status (SES), Mini-Mental State Examination (MMSE), Clinical Dementia Rating (CDR), Estimated Total Intracranial Volume (eTIV), Normalized Whole Brain Volume (nWBV), and Atlas Scaling Factor (ASF). Prior work in the field is referenced, highlighting studies that predominantly focused on raw MRI data analysis. In contrast, this study introduces a unique approach by utilizing a curated set of biomarkers, allowing for a more targeted and potentially interpretable model. Machine learning models such as Logistic Regression, Support Vector Machine, Decision Tree, Random Forest Classifier, and AdaBoost are employed, with performance measured using established metrics. Information about severity and state are stored during the EADDLS module and used for ADmod. ADmod uses the stored MRI data during the EADDLS module to model the growth of amyloid {beta} build-up in the brain using convolution, resulting in both generalizable approaches and patient-specific approaches. There have been numerous mathematical instantiations to model amyloid {beta} build-up using partial differential equations (or PDEs), these however have remained unincorporated due to prolonged runtimes and storage limitations along with those of pre-set conditions. We propose a novel amyloid {beta} growth model using deep encoder-decoder networks in conjunction with convolution. The study contributes to the growing body of research in early Alzheimers detection, offering insights, results, and a discussion of limitations. The conclusion outlines a unique approach, emphasizes the practical implementation of the proposed model, acknowledges limitations, and suggests avenues for further research. Early detection of AD can significantly better the patients quality of care and lead to future preventative or risk assessment measures.