Paris Alexandros Lalousis

Elevated pro-inflammatory markers have been reported in schizophrenia, and inflammation has been found to modulate brain functional connectivity. This paper provides novel evidence on the association between striatal connectivity and inflammation in first-episode schizophrenia (FES) and its relationship with negative symptoms and cognitive performance. We included 132 individuals with FES who underwent resting-state functional MRI (fMRI) as part of the Benefit of Minocycline on negative symptoms of schizophrenia (BeneMin) trial. We applied a previously estimated semi-supervised machine learning clustering solution on inflammatory patterns to assign individuals with FES to an inflammation group at baseline and at 12-month follow-up. A seed-based functional connectivity analysis was conducted using bilateral basal ganglia as seed regions. Symptoms of psychosis were evaluated using the Positive and Negative Syndrome Scale (PANSS). The Digit Symbol Substitution Test (DSST) was used as a proxy for cognitive function. At baseline, FES individuals in the elevated CRP group (High-CRP) presented higher functional connectivity between the right nucleus accumbens and the left middle frontal gyrus and between the right pallidum and the left middle frontal gyrus/precentral gyrus, compared with individuals in the low inflammation group (Low-Inf). These differences were also evident at follow-up, although they did not reach statistical significance. We found a greater correlation between right nucleus accumbens—left superior and middle frontal gyri functional connectivity and PANSS negative symptom subscale scores in the High-CRP group compared with the Low-Inf group. Our findings suggest that inflammation is associated with differences in fronto-striatal connectivity early in schizophrenia, which could, in turn, be associated with more severe negative symptoms. Greater fronto-striatal connectivity may represent a distinct neural signature associated with greater inflammation. Interventions aiming at reducing inflammation may represent novel strategies for modulating striatal connectivity and ameliorating negative symptoms.

Predicting individual mental-health trajectories requires models that capture cross-domain dynamics rather than only statistical associations. We introduce the Coupled Stochastic Dynamical System (CSDS), a forward generative model of interacting digital behaviour, wearable-derived physiology and psychiatric burden. In the multi-year GLOBEM student cohort (N=496), Bayesian inversion recovered four behavioural phenotypes, reproducing 80–83% of cluster targets through distinct trait-vulnerability, physical-activity and affective-response parameter combinations. These phenotypes were primarily differentiated by digital-media overuse, physical activity, sleep and affective burden. Participant-specific inversion then generated virtual twins whose synthetic trajectories preserved observed behavioural traces: hidden Markov models trained on virtual-twin data reproduced state occupancy and transition patterns ( r =0.84–0.87 and 0.93–0.94). Forecasts outperformed causal mean and persistence baselines for short-horizon magnitude prediction and predicted affective-state changes with 66–78% balanced accuracy over 21 days. These findings establish CSDS as an interpretable framework for simulating individual behavioural dynamics and advancing personalised mental-health modelling.

Background: Major depressive disorder (MDD) is clinically heterogeneous, hindering identification of reproducible biomarkers. Using a semi-supervised machine learning approach, HYDRA, we previously identified two neuroanatomical dimensions from structural MRI in medication-free MDD from COORDINATE-MDD consortium. These dimensions (D1, D2) showed differential responses to selective serotonin reuptake inhibitor (SSRI) antidepressants and placebo. External replication in UK Biobank linked D2, characterized by widespread subtle neuroanatomical reductions, to an immuno-metabolic profile. Here, we examined whether these dimensions are detectable early in the course of illness. Methods: We applied the pre-trained model to structural MRI data from the multisite PRONIA cohort, comprising individuals with recent-onset depression (ROD; n = 377; mean age 25.8 years, SD 6.0; 51.3% female) and healthy controls (n = 267; mean age 25.5 years, SD 6.4; 61.0% female). Participants were assigned to clusters (C1, C2) corresponding to the previously identified dimensions (D1, D2). Clusters were compared on clinical symptom profiles, peripheral inflammatory markers, and in a subset (n = 107), proteomic ageing indices. Results: Two neuroanatomical clusters were identified in PRONIA. C1 (n = 265) showed higher negative symptom severity and elevated interleukin-2 levels. C2 (n = 140) was associated with higher residual proteomic age. Overall depressive symptom severity did not differ significantly between clusters. Conclusions: Neuroanatomical dimensions of MDD are reproducible and detectable at illness onset. Associations with negative symptom severity, inflammatory signalling, and proteomic ageing suggest these dimensions capture biologically meaningful heterogeneity early in depression. These findings support a biologically informed framework for stratified treatment approaches in MDD.

Ketamine has demonstrated rapid antidepressant efficacy in treatment-resistant depression (TRD), but clinical decision-making is challenging due to variability in individual response. Current trial-and-error prescribing practices may expose patients to ineffective treatment and avoidable adverse effects, underscoring the need for reliable predictive tools to optimize treatment selection and support personalized, evidence-based care. We developed a machine-learning model (support vector classifier) to predict antidepressant response to ketamine using pre-treatment structural MRI data. The model was trained on 99 adults with TRD given a single intravenous ketamine infusion (0.5 mg/kg). Clinical response was defined as a ≥50% reduction in MADRS scores 24 h post-infusion. Internal validation used repeated nested cross-validation, and generalizability was tested in two independent ketamine-treated cohorts ( n = 51) and a saline-treated control group ( n = 49). Among ketamine-treated participants, 52 (52.5%) responded to treatment. The model achieved a balanced accuracy of 72.2% (sensitivity = 72.3%, specificity = 73.1%, AUC = 0.72) in the discovery sample and 60.0% ( p = 0.01, AUC = 0.65) in external validation. Greater gray matter volume in frontal regions predicted response, whereas greater cerebellar volume predicted non-response. Performance dropped to chance in the saline cohort (BAC = 41.1%, AUC = 0.45), supporting pharmacologic specificity. These findings present the first machine-learning model for the prediction of ketamine response in TRD using structural neuroimaging and highlight its potential utility for stratified treatment planning and biomarker-informed interventions while providing mechanistic insight into neuroanatomical predictors of antidepressant response.

Some schizophrenia patients share characteristics with behavioral variant frontotemporal dementia (bvFTD) including gray matter volume (GMV) similarities, suggesting overlapping brain mechanisms that may contribute to disease heterogeneity in schizophrenia. However, it is not yet understood whether schizophrenia patients with a bvFTD-like GMV signature also show additional features of bvFTD. Seventy-six patients with schizophrenia (mean age 31.2 years; 71% male) from a previous clinical trial (NCT04038957) underwent structural magnetic resonance imaging (MRI) to measure GMV. Healthy controls ( n = 79) were used for age-related dynamic standardisation. Subsets of patients also completed [¹⁸F]-DOPA positron emission tomography (PET; n = 40) for striatal dopamine synthesis, neuromelanin-sensitive MRI (NM-MRI; n = 68) for midbrain dopaminergic integrity, and quantitative susceptibility mapping (QSM; n = 69) for brain iron accumulation. A previously validated machine-learning–based GMV classifier quantified bvFTD-like pattern expression in each schizophrenia patient. Associations with neurochemical, clinical, and medication measures were examined using correlation analyses. Higher bvFTD scores were associated with lower striatal dopamine synthesis capacity ( r = –0.343, p = 0.032) and higher striatal QSM values ( r = 0.282, p = 0.020), but showed no significant association with QSM or NM-MRI values in the dopaminergic midbrain ( p = 0.903 and p = 0.102, respectively). No significant associations were found with negative symptom severity or with medication. Schizophrenia patients expressing a stronger bvFTD-like GMV pattern show lower striatal dopamine synthesis and elevated striatal iron, both hallmark features of bvFTD. This contrasts with the hyperdopaminergic model of schizophrenia and suggests distinct, potentially neurodegenerative mechanisms.

The transition from narrow AI to Artificial General Intelligence (AGI) presents a shift that necessitates moving beyond purely algorithmic solutions toward a sociotechnical perspective. Our article argues that determining the moral compass of AGI is fundamentally a problem of cognitive compatibility and design constraints. We integrate established Human Factors and Ergonomics (HFE) principles, specifically Cognitive Work Analysis (CWA), Joint Cognitive Systems, and Distributed Situation Awareness, to address the process gulf arising from the opacity of AGI goal pursuit. We critically examine seminal AI safety thought experiments through the lens of automation surprises and mode error. Furthermore, we adapt Rasmussen’s Risk Management Framework to model AGI alignment as a hierarchical control problem, introducing the concept of Ecological Alignment Interfaces to visualise the safety envelope. Finally, we argue that AGI safety requires Bidirectional Human-AI Alignment and the rigorous, non-iterative safety protocols found in High Reliability Organisations (HROs). We conclude that the HFE discipline offers the essential toolkit for defining the design constraints necessary for safe AGI.Practitioner Summary:This article argues that Artificial General Intelligence (AGI) is a sociotechnical challenge requiring Human Factors expertise. This article demonstrates how practitioners can apply Cognitive Work Analysis and Systems Theoretic Accident Model and Processes to define the moral compass of AGI as a set of rigorous design constraints. We provide a framework for using Rasmussen’s Risk Management model to design governance structures that maintain human control over opaque, autonomous systems.

In this study, we present and evaluate AIM Review Tool, a modern web-based application that combines active and supervised machine learning strategies to accelerate the screening of publications for systematic reviews. AIM Review incorporates state-of-the-art text vectorization methods and machine learning models implemented directly on the web browser. Unlike existing tools, it is the first application to integrate nested-cross-validated and semi-automated methods for systematic review screening, enhancing both efficiency and precision of paper selection, thus accelerating evidence synthesis. Using three real-world case studies, we tested the capabilities of both active learning and supervised learning pipelines. Active learning prioritization of relevant studies showed a reduction in screening workload (i.e., percentage of publications that did not need to be screened) from 92% to 20% depending on the characteristics of the systematic review. Supervised learning pipelines using a subset of the screenings predict the relevance of unscreened publications with balanced accuracies between 75% and 87%, using nested cross-validation. AIM Review offers a flexible, efficient solution for large-scale literature screening, allowing for seamless integration with the manual review processes. Future work will focus on automating subsequent stages, including data extraction, to further improve the speed and scalability of systematic reviews.

Biological risk signatures could aid the early detection of schizophrenia, but their precision likely depends on the clinical risk definitions they are derived from. Using machine learning, we analyzed structural MRI data from 1,425 patients and 907 healthy individuals in a multi-site and multi-diagnostic database to detect and validate signatures of different risk syndromes―Cognitive Disturbances (COGDIS), Ultra-High-Risk (UHR) or COGDIS + UHR―compared to schizophrenia. Patients with COGDIS but not UHR-related syndromes were detectable using MRI. COGDIS and schizophrenia brain signatures were highly correlated based on shared prefronto-parieto-perisylvian volume reductions. The expressions of these brain signatures intensified from healthy individuals through affective disorders and non-schizophrenic psychoses to patients with schizophrenia, and they could be predicted with up to 21% variance explained by genetic, neurocognitive, and phenotypic risk factors. COGDIS and schizophrenia signature expressions predicted poor functional outcomes after two years. These results emphasize the need to refine early detection tools by integrating cognitive basic symptoms and their neural correlates into schizophrenia risk assessment.