Multimodal AI for Cognitive Screening and Mri-Based Classification

This paper proposes a multimodal artificial-intelligence framework for preliminary cognitive screening and research-oriented MRI classification. The planned framework brings together fifteen cognitive games organized into five groups, speech/audio analysis, eye-movement tracking, and an MRI-based deeplearning component. The five target categories are mild cognitive impairment (MCI), Alzheimer’s disease, attention-deficit/hyperactivity disorder (ADHD), Parkinson’s disease, and dementia, with three proposed cognitive games associated with each category. The game groupings reflect the project’s design organization and should not be interpreted as proving that any game is specific to, or diagnostic of, a particular condition. The system is intended to collect behavioural observations and, if supported by suitable datasets, examine whether MRI-derived features can contribute to classification. The integrated multimodal model, MRI pipeline, and clinical performance have not been verified as completed. Earlier project work included a prototype with a subset of the cognitive games and backend data storage; the status of each feature must be checked against the current repository before publication. This paper therefore presents the proposed architecture, planned methodology, and evaluation protocol rather than claiming experimental results. A key methodological concern is that MCI, Alzheimer’s disease, and dementia are overlapping clinical concepts and cannot automatically be treated as mutually exclusive classes. Dataset labels, task definitions, and clinical interpretation must be specified before model training.