MTWIN: A Mental Health Monitoring Framework toward a Digital Twin

Abstract

Digital Twins (DTs) are emerging as a transformative technology across industries, offering a virtual representation of physical objects, including machines, cars, industrial robots, processes, and even individuals. The present work proposes a DT-inspired mental health monitoring framework, MTWIN, that aggregates personal and medical information, excluding sensitive visual data such as facial photographs or medical imagery. Our work provides healthcare professionals with a tool to monitor and evaluate the patient's psychological progress throughout treatment. Data sources for MTWIN include manually entered information (e.g., diary entries, medication intake, eating habits, social interactions, and self-reported general health status) and sensor data from wearable devices (e.g., physical activity, sleep duration), combined with weather data based on the patient's location. Additionally, MTWIN integrates predictions from a Machine Learning (ML) model trained on the TREMO dataset (a validated emotion analysis dataset in Turkish, with 4709 participants) and the Turkish Tweets dataset. Based on diary entries, the model analyzes emotions such as joy, sadness, fear, anger, surprise, and disgust. MTWIN is visualized through mobile, tablet, and web applications, providing patients with feedback to improve their daily physical activity, sleep duration, and eating habits, while offering healthcare professionals a historical view of their mental health progression. Preliminary prototype simulations indicate that the framework can successfully aggregate and visualize the data necessary to track a patient's response to medication and treatment. The ML model, DistilBERTurk fine-tuned with 23,462 entries, achieves 93.5% accuracy on the validation dataset.