AI-Based Handwriting Analysis for Early Detection of Developmental and Cognitive Disorders

Developmental and Cognitive Disorders, encompassing both adult-onset neurological conditions like Alzheimer's disease, Parkinson's disease, and Lewy body dementia, and pediatric conditions such as specific learning disorders, present a growing global health challenge. These conditions often progress, causing irreversible damage before symptoms become apparent, highlighting the urgent need for early and accurate diagnostic tools. This thesis examines how artificial intelligence methods, particularly quantitative handwriting analysis, can aid in the early detection and ongoing monitoring of these disorders. Handwriting, a complex integration of motor and cognitive functions, serves as a sensitive indicator of neurological health. Changes in handwriting characteristics, including pressure, speed, and stroke patterns, often emerge early in the course of neurological disorders and exhibit distinct features in specific learning disorders. This work introduces new ways to utilise these changes. For Alzheimer's disease, the thesis develops artificial intelligence-based handwriting analysis techniques to pinpoint diagnostically relevant features and integrate them within machine learning frameworks, comparing their effectiveness with traditional methods. These approaches include analysing handwriting at the stroke level for fine-grained insights, employing multimodal artificial intelligence that combines both manually engineered features and features extracted through deep learning, and using explainable artificial intelligence techniques to ensure results are clear and clinically meaningful. The research also examines how cognitively demanding writing tasks affect diagnostic accuracy. This thesis initiates a longitudinal study focused on specific learning disorders, comprising an extensive four-year data collection effort designed to establish a complete dataset of handwriting. A preliminary analysis of the collected data examines potential handwriting indicators for the early identification of specific learning disorder. This analysis also assesses whether artificial intelligence methods, typically applied to adult neurological disorders, can be adapted for pediatric neurodevelopmental conditions. This study demonstrates the adaptability of artificial intelligence-driven handwriting analysis methods across different types of developmental and cognitive disorders. Furthermore, this thesis addresses practical issues involved in moving these methods from research settings to clinical practice. This includes managing variations in writing tools and individual writing styles. By creating standardised, non-invasive, and cost-effective handwriting-based assessments, this research seeks to improve access to neurological evaluations, particularly in communities with limited resources. The potential influence extends to supporting long-term disease tracking, identifying biomarkers specific to different populations, and promoting timely interventions during critical developmental periods or disease stages when treatments are most effective. This ultimately aims to improve patient outcomes and lessen the societal impact of developmental and cognitive disorders. This work establishes artificial intelligence-powered handwriting analysis as a promising approach to make early detection a regular part of managing developmental and cognitive disorders.