A Network Analysis Approach for the Classification of Psychiatric Disorders Using Multi-Modal Data

Psychiatric disorders span a broad range of conditions, exhibiting diverse clinical and behavioral traits. Traditional diagnostic methods primarily reliant on subjective behavioral assessments present substantial challenges in achieving accurate identification and characterization. To overcome this challenge, we propose an innovative data-driven integrated classification approach that combines mobility data from wearable sensors with clinical data. Initially, we construct network representations of mobility patterns, enabling the identification of distinct communities within the population. Subsequently, we classify each community by exploring its unique clinical parameters. Our results reveal the presence of at least five distinct communities within the studied population, each characterized by a unique combination of mobility and clinical data. These groups are identified as follows: ADHD (Attention-Deficit / Hyperactivity Disorder) with ADD (Attention-Deficit Disorder), Bipolar with Anxiety, Unipolar with Anxiety, and ADHD with Bipolar and Anxiety. This classification is achieved by leveraging both mobility and clinical data present in the dataset, allowing for a more comprehensive and nuanced understanding of these psychiatric conditions. The resulting findings provide valuable insights into the heterogeneity associated with different disorders, thereby facilitating the development of personalized interventions and treatment strategies. The salient aspect of this study is that our results are neither predicated solely on class labels nor driven solely by mobility data; instead, they derive strength from the harmonious integration of data-driven analysis and crucial clinical insights, culminating in a robust classification model.