Machine learning algorithm in predicting Alzheimer’s disease: exploring brain volumetric markers and cognitive profiles in amnestic mild cognitive impairment patients

The increasing number of people diagnosed with Alzheimer’s disease (AD) represents one of the most pressing challenges facing contemporary healthcare, prompting a surge of research aimed at understanding its progression and finding effective interventions. Since pharmacological therapies for cognitive and behavioral treatments of AD symptoms have shown poor efficacy, preventive actions, by targeting risk factors and promoting healthy lifestyles, actually represent the first option to manage such symptoms. Mild Cognitive Impairment (MCI), as an intermediate stage between normal cognition and dementia, serves as a critical juncture in the spectrum of cognitive decline and is conceived as a potential target for interventions aimed at delaying dementia progression and mitigating associated disabilities. The feasibility of intervening from the earliest stages is even more important in light of the growing evidence showing that not all patients with MCI progress to AD and, for this reason, MCI presents a unique opportunity to investigate factors that may influence the AD conversion. Consequently, a significant body of research has focused on neuroimaging and neuropsychological features that distinguish MCI patients who progress to AD from those remaining clinically stable in defined time windows, thus contributing to the theoretical frameworks that strengthen our understanding of the disease. Among several approaches used in the neurodegenerative field, machine learning (ML) algorithms have recently emerged as powerful and promising tools, offering unprecedented opportunities in the analysis of complex datasets and providing clinicians with vital information that can guide treatment decisions and patient management strategies. In the context of Alzheimer’s research, several studies demonstrated the utility and the accuracy of ML methods in identifying predictors of AD conversion, thereby offering prognostic measures contributing to determine patterns within neuroimaging and cognitive assessment data that may be indicative of conversion risk. This thesis explores the transition from MCI to AD through the lens of machine learning, and specifically it aims at identifying the brain-volumetric predictors of AD conversion and at exploring neuropsychological profiles of amnestic MCI patients converting to possible AD and of those remaining stable within a one-year period. Through a comprehensive examination of cerebral regions recognized to be involved in AD, we discern biomarkers indicative of an elevated likelihood of disease progression, facilitated by the novel application of feature selection methodology that enhance our models by differentiating the most pertinent anatomical alterations associated with disease conversion. Analysis of neuropsychological differences and the impact of static cognitive reserve on performances also provides a behavioral context, thereby revealing the complex interactions between brain volumes, cognitive reserve and cognition. By integrating the innovative aspect of AI tools and the potential beneficial effects of cognitive reserve, this dissertation thus aims to contribute to the growing body of knowledge pertaining to AD, providing healthcare practitioners with the requisite instruments for prompt diagnosis and intervention. This endeavor is particularly critical within a context wherein timely interventions and accurate diagnoses have the potential to modify disease trajectories and enhance patient outcomes.