Evaluating progress throughout a patient's rehabilitation episode is critical for determining the effectiveness of the selected treatments and is an essential ingredient in personalised and evidence-based rehabilitation practice. The evaluation process is complex due to the inherently large human variations in motor recovery and the limitations of commonly used clinical measurement tools. Information recorded during a robot-assisted rehabilitation process can provide an effective means to continuously quantitatively assess movement performance and rehabilitation progress. However, selecting appropriate motion features for rehabilitation evaluation has always been challenging. This paper exploits unsupervised feature learning techniques to reduce the complexity of building the evaluation model of patients' progress. A new feature learning technique is developed to select the most significant features from a large amount of kinematic features measured from robotics, providing clinically useful information to health practitioners with reduction of modeling complexity. A novel indicator that uses monotonicity and trendability is proposed to evaluate kinematic features. The data used to develop the feature selection technique consist of kinematic data from robot-aided rehabilitation for a population of stroke patients. The selected kinematic features allow for human variations across a population of patients as well as over the sequence of rehabilitation sessions. The study is based on data records pertaining to 41 stroke patients using three different robot assisted exercises for upper limb rehabilitation. Consistent with the literature, the results indicate that features based on movement smoothness are the best measures among 17 kinematic features suitable to evaluate rehabilitation progress.

Gao et al. introduced a deep learning framework for health risk assessment in adolescents by developing a combination of a two-dimensional convolutional autoencoder (2DCNN-AE) with multisequence learning and multi-scale asynchronous correlation information extraction techniques. Applied to modelling fMRI data from the Adolescent Risk Behavior (AHRB) dataset, the framework was designed to capture both spatial and time-shifted relationships among brain regions considering limited labelled data. The study demonstrated the promising performance of a deep learning-based approach for understanding and predicting health-related risks in adolescents. The work illustrated the potential of the developed deep learning method in enhancing the precision of health risk assessment and supporting early identification of individuals who may benefit from timely intervention. However, more robust external validation and assessment of generalisability across diverse populations and settings are essential to establish the clinical utility and readiness for real-world implementation.Chalmers et al. developed a study of using deep reinforcement learning to simulate the effect of dendritic spine loss as observed in major depressive disorder (MDD). The simulated spine loss produced a variety of behaviours in the artificially intelligent agent, such as anhedonia, increased temporal discounting, avoidance, and an altered exploration/exploitation balance. Furthermore, simulating alternative and longstanding reward-processing-centric conceptions of MDD (dysfunction of the dopamine system, altered reward discounting, context-dependent learning rates, increased exploration) did not produce the same range of MDD-like behaviours. This study highlighted the potential of AI as a powerful tool for investigating underlying mechanisms beyond diagnosis and outcome prediction, providing a computational approach to probe the pathophysiology of MDD and inform hypotheses for potential therapeutic strategies.Haykal et al. contributed a perspective in examining opportunities, limitations, and ethical considerations of leveraging AI in detecting body dysmorphic disorder (BDD), a frequently overlooked yet clinically significant condition. BDD characterised by a persistent and intrusive preoccupation with perceived flaws in appearance, typically minor or imperceptible to others. Individuals with BDD often seek repeated cosmetic procedures in hopes of achieving relief, but they typically find little psychological benefit from such interventions. In this perspective, the authors explored the potential role of using AI approaches in improving the management of BDD, such as early diagnosis, prompt reflection, highlight concerning patterns, and enhancing clinical sensitivity, considering transparency, empathy, and restraint.Human-machine interaction and computational neuroscience also provide promising opportunities to advance rehabilitation, particularly for older adults (Lu et al. 2021). In the work from Xing et al., a deep learning model was developed to process point-cloud data acquired from laser sensors for non-contact analysis of hand gestures and surface features, with the aim of supporting hand function rehabilitation with human-machine interaction. By analysing hand surface point clouds, local feature extraction, and abstraction and enhancement of dimensional information, the proposed approach effectively captured relevant spatial information and demonstrated promising performance in gesture and handsurface feature recognition. These findings highlight the potential of combining AI with non-contact sensing technologies to support more natural, accessible, and data-driven approaches to rehabilitation and human-machine interaction.Finally, Liu et al. reviewed the application of AI-based radiogenomics in managing neurodegenerative diseases, such as Multiple Sclerosis (MS), Dementia, Parkinson's Disease (PD), Amyotrophic Lateral Sclerosis (ALS), and Creutzfeldt-Jakob Disease. The study summarised representative AI models in neurodegenerative diseases, a range of publicly available data resources, supporting radiomics, genomics, and radiogenomics studies. It also highlighted key directions for future research in computational modelling for neurodegenerative diseases, including strategies to mitigate overfitting in high-dimensional genomic data, robust validation approaches that account for modality-specific variability and incomplete data, and efforts to bridge the gap between AI-enhanced models and their translation into real-world clinical practice.Looking forward, health data science and AI are rapidly evolving fields, with emerging methodologies providing opportunities beyond those demonstrated in this collection. For example, large language models may enable more sophisticated analysis and interpretation of complex information, supporting personalised risk assessment and tailored intervention (Li et al. 2026), while multimodal learning could integrate complementary data sources to generate more comprehensive representations of individual health (Liu et al. 2025). Integrating subjective (e.g., questionnaires) and objective measurements (e.g., wearable sensing data) may further enhance the understanding of individual health (Lim et al. 2024). Further advances in model optimisation, robustness, interpretability, and clinical validation will be essential for translating these innovations into real-world applications (Shen et al. 2024). Ultimately, interdisciplinary synergy across neuroscience, psychology, clinical medicine, and computational sciences will be critical to developing AI systems that are technically capable, clinically meaningful, and generalisable across diverse populations and settings. We hope this Research Topic will stimulate further interdisciplinary research and responsible translation of health data science and AI in neuroscience and psychology.

Evaluating progress throughout a patient’s rehabilitation episode is critical for determining effectiveness of the selected treatments and contributing to the evidence-based practice. The evaluation process is complex due to the inherent large human variations in motor recovery and the limitations of commonly used clinical measurement tools. Information recorded during a robot-assisted rehabilitation process can provide an effective means to continuously quantitatively assess movement performance and rehabilitation progress. However, selecting appropriate motion features for rehabilitation evaluation has always been challenging. This paper exploits unsupervised feature learning techniques to reduce the complexity of building the evaluation model of patients’ progress. A new feature learning technique is developed to select the most significant features from a large amount of kinematic features measured from robotics, providing clinically useful information to health practitioners with reduction of modeling complexity. A novel indicator that can reflect monotonicity and trendability is proposed to evaluate the suitability of kinematic features, which are derived from the collected data of a population of stroke patients participating in robot-aided rehabilitation. The selected kinematic features allow for human variations across a population of patients as well as over the sequence of rehabilitation sessions. The study is based on data records pertaining to 41 stroke patients using three different robot assisted exercises for upper limb rehabilitation. Consistent with the literature, the results indicate that features based on movement smoothness are the best measures among 17 kinematic features used to evaluate rehabilitation progress.