Sophie Tascedda ¹ Pierfrancesco Sarti ^2,3* Veronica Rivi ² Claudia Savia Guerrera ⁴ Giuseppe Alessio Platania ⁴ Mario Santagati ⁵ Filippo Caraci ^6,7 Johanna MC Blom ^2,8*

• ¹ Department of Advanced Computing Sciences, Faculty of Science and Engineering, Maastricht University, Maastricht, Netherlands, Netherlands
• ² Department of Biomedical, Metabolic and Neural Sciences, University of Modena and Reggio Emilia, Modena, Emilia-Romagna, Italy
• ³ Psychiatric University Hospital Zurich, Zurich, Zürich, Switzerland
• ⁴ Department of Education Sciences, University of Catania, Catania, Sicily, Italy
• ⁵ Azienda Sanitaria Provinciale di Catania, Catania, Sicily, Italy
• ⁶ Department of Drug and Health Sciences, University of Catania, Catania, Sicily, Italy
• ⁷ IRCCS Oasi Maria SS, Troina, Italy
• ⁸ Centre for Neuroscience and Neurotechnology, University of Modena and Reggio Emilia, Modena, Italy

Background: Alzheimer's disease and mild cognitive impairment are often difficult to differentiate due to their progressive nature and overlapping symptoms. The lack of reliable biomarkers further complicates early diagnosis. As the global population ages, the incidence of cognitive disorders increases, making the need for accurate diagnosis critical. Timely and precise diagnosis is essential for the effective treatment and intervention of these conditions. However, existing diagnostic methods frequently lead to a significant rate of misdiagnosis. This issue underscores the necessity for improved diagnostic techniques to better identify cognitive disorders in the aging population. Methods: We used Graph Neural Networks, Multi-Layer Perceptrons, and Graph Attention Networks. GNNs map patient data into a graph structure, with nodes representing patients and edges shared clinical features, capturing key relationships. MLPs and GATs are used to analyse discrete data points for tasks such as classification and regression. Each model was evaluated on accuracy, precision, and recall.The AI models provide an objective basis for comparing patient data with reference populations. This approach enhances the ability to accurately distinguish between AD and MCI, offering more precise risk stratification and aiding in the development of personalised treatment strategies.The incorporation of AI methodologies such as GNNs and MLPs into clinical settings holds promise for enhancing the diagnosis and management of Alzheimer's disease and mild cognitive impairment. By deploying these advanced computational techniques, clinicians could see a reduction in diagnostic errors, facilitating earlier, more precise interventions, and likely to lead to significantly improved outcomes for patients.