Computational Design of Therapeutic Digital Environments: A Deep Learning Approach to Personalized Mental Well-being Intervention

INTRODUCTION: The global rise of mental health challenges underscores the urgent need for personalized and non-pharmacological interventions. However, current digital mental health tools often depend on generalized design principles that overlook individual differences in aesthetic preference and affective response. This gap limits long-term engagement and reduces the effectiveness of affective regulation. To overcome these constraints, this study explores the integration of art-psychology principles with advanced machine learning techniques to create adaptive therapeutic environments. OBJECTIVES: The objective of this paper is to develop and evaluate a novel Computational Design Framework (CDF) capable of generating personalized Therapeutic Digital Environments (TDEs) through real-time affective feedback, thereby improving both user experience and therapeutic efficacy. METHODS: The proposed framework combines deep learning–based aesthetic generation with dynamic environment optimization. A Generative Adversarial Network (GAN) is used to produce personalized visual and auditory stimuli, while a Physiological Signal Processing (PSP) module analyzes real-time biosignals—including heart rate variability and skin conductance—to infer users’ affective states. A Deep Reinforcement Learning (DRL) model then adjusts TDE parameters based on both physiological and self-reported feedback. A controlled experiment involving 50 participants was conducted to evaluate the framework against static, generalized TDEs. RESULTS: The DRL-optimized TDEs achieved a 25.3% greater reduction in physiological stress markers compared to static TDEs and yielded higher user satisfaction. Analysis revealed key design parameters—such as specific ranges of color saturation and sound frequency bands—that consistently correlated with positive affective shifts. The findings indicate the framework’s capability to identify and personalize aesthetic variables that influence emotional regulation within the current experimental scope. CONCLUSION: This research establishes a replicable, data-driven methodology for designing therapeutic interventions that bridge subjective aesthetic experience with objective physiological outcomes. The proposed CDF advances cross-disciplinary innovation at the intersection of art, psychology, and technology, suggesting promising directions for personalized healthcare and computationally driven design practices.