The use of ketamine as a treatment for mood disorders continues to grow in popularity. However, the differences between individuals consuming ketamine within clinical settings and those self-medicating in casual, non-clinical settings has not been yet characterized. Using the 2023 Global Psychedelic Survey, this study aimed to examine the demographic and clinical characteristics, as well as the self-reported effectiveness and hallucination-related side effects, of ketamine use by comparing individuals using ketamine with and without medical supervision. Compared to clinical ketamine consumers (n = 238), non-clinical consumers (n = 222) were more experienced in use of various psychedelic (p < .0001) and non-psychedelic recreational substances (p < .001), and lifetime use of ketamine (p < .0001). Both groups presented with similar mild depression and anxiety symptoms at the time of the survey collection (p = .22, p = .46). Ketamine was similarly and highly rated in effectiveness within both groups when rated on a scale of 0-100 (83.3 ± 15.0 clinical, 81.2 ± 16.6 non-clinical; p = .45). Non-clinical consumers presented higher odds of reporting long-term hallucinations following ketamine use (OR = 2.58, 95% CI: 1.08-6.31). The results of this study indicate a high self-reported effectiveness of ketamine in clinical and non-clinical settings while emphasizing the potential risks of side effects when used without medical supervision.

This essay explores the role of visual images in dream imagery as a space where unconscious processes intersect with historical forces. It offers a comparative reading in Freud's Interpretation of Dreams (1900) and Benjamin's Arcades Project (1982). Freud conceptualizes dream images as condensed memory traces that navigate between perception and hallucination, revealing repressed desires. Conversely, Benjamin's dream images critique modernity's loss of aura, offering moments of "Now-Time" as opportunities for historical creation. The essay argues that both Freud and Benjamin share a foundational premise: turning a blind eye to the obscure and unknown dimensions of individual and collective experience fosters crises and destruction, whereas striving to reveal these hidden areas promotes transformation. Exploring disruptions and ruptures in the continuity of experience challenges the inclination to act according to Freud's concept of repetition compulsion or Nietzsche's eternal return, as interpreted by Benjamin. These ideas are explored through the dream of the burning child presented by Freud and the ghost dream presented by Benjamin. The dialectic of insight and blindness, alongside the image of open doors in both dreams, invite us to consider how, even amidst loss and rupture, dream images can open pathways to new realms of experiencing and remembering.

BackgroundNeuropsychiatric symptoms (NPS) in Alzheimer's disease (AD) often occur in clusters, and their clinical meaning may depend on the symptoms with which they co-occur. Whether a previously identified NPS cluster pattern remains transportable after expansion of an amyloid PET-positive cohort is uncertain.ObjectiveTo assess the transportability of a previously published NPS cluster model and to characterize symptom-role reorganization after cohort expansion.MethodsWe analyzed 301 consecutive drug-naive patients with probable AD and positive 18F-FC119S amyloid PET. The cohort comprised the previously published 143-patient reference cohort and 158 newly enrolled, non-overlapping patients who constituted the extension cohort. Twelve Korean Neuropsychiatric Inventory domains were coded as present or absent. Hierarchical clustering was repeated in the extension cohort and in the full cohort, and changes in individual symptom relationships were examined.ResultsThe original four-cluster solution was only partly reproduced. Aggression, disinhibition, and aberrant motor behavior remained closely grouped, while anxiety and irritability also remained linked and were joined by euphoria. Hallucination shifted into the behavioral-dyscontrol cluster, whereas depression, apathy, nighttime behavior, and eating abnormality formed a vegetative-apathy cluster. Delusion formed a single-domain cluster and, together with aggression, aberrant motor behavior, and disinhibition, showed the largest stage-associated prevalence differences between Clinical Dementia Rating (CDR) 0.5 and CDR 1.0-2.0.ConclusionsNPS clusters in amyloid PET-positive AD may not represent fixed syndromes. A symptom-role framework that distinguishes preserved relationships, reorganized symptoms, and independent stage-linked symptoms may provide a more clinically informative understanding of NPS in AD.

Outpatient clinic letters are a cornerstone of clinical communication but frequently exceed recommended reading levels, limiting patient comprehension and engagement. Large Language Models (LLMs) and Artificial Intelligence (AI) have been proposed as scalable tools for generating and simplifying these documents, but the evidence regarding their clinical accuracy, safety, and effectiveness remains fragmented. We conducted a systematic review of studies evaluating AI or LLMs for generating, simplifying, or enhancing outpatient clinic letters. Five databases (PubMed, EMBASE, Web of Science, CENTRAL, and CINAHL) were searched from inception to 1 November 2025. Two independent reviewers screened studies, extracted data, assessed quality using the Mixed Methods Appraisal Tool (MMAT), and certainty using the GRADE appraisal tool. Seven studies were included, comprising two studies using real-world clinic data and five using synthetic or hypothetical scenarios. Findings regarding readability were mixed; while some AI models improved readability scores compared to human-authored letters, most AI-generated content still failed to meet the recommended US Grade 6 reading level. The two studies that measured patient-reported outcomes reported high satisfaction and comprehension with AI-simplified letters, but these studies relied on subjective measures of understanding or lacked a human-authored control group. Clinical accuracy varied substantially, with information fidelity ranging from 10% to 100%. Risks of 'hallucinations', inappropriate medical advice, and tactless patient descriptions were documented. One study reported a ten-fold reduction in drafting time compared to human dictation. Overall certainty of evidence was rated very low across all primary outcomes due to methodological limitations, small sample sizes, and reliance on hypothetical scenarios. The current evidence supporting AI and LLMs use for outpatient clinic letters is preliminary and limited, providing insufficient evidence to support routine clinical implementation beyond supervised experimental and quality-improvement settings. Concerns regarding accuracy, safety, and generalisability remain. Further real-world patient-centred research is required before widespread adoption. PROSPERO registration: CRD420251181303.

Recent advances in ophthalmic foundation models have accelerated the application of artificial intelligence in ophthalmology, improving disease detection, progression assessment, and treatment evaluation. These advances are supported by the increasing availability of multimodal ophthalmic imaging data, including optical coherence tomography, color fundus photography, and slit lamp imaging, together with progress in self-supervised learning, vision-language modeling, and agent-based systems. This review summarizes publicly available multimodal ophthalmic imaging datasets and examines their foundational role in large-scale pretraining and cross-modal representation learning. We organize ophthalmic foundation models into vision foundation models, vision-language foundation models, and Large Language Model-based agentic systems, and discuss their core architectures, learning strategies, and representative applications. By synthesizing evidence across structural, vascular, and anterior-segment modalities, we illustrate how ophthalmic artificial intelligence is evolving from task-specific image recognition toward semantically grounded, clinically oriented cognition and interactive reasoning. We further highlight key challenges that hinder clinical translation, including data imbalance across modalities, imperfect image-text semantic alignment, limited robustness and generalizability, hallucinations in generative agents, and barriers to clinical workflow integration. Finally, we outline future directions, emphasizing standardized multimodal data resources, improved interpretability and factual consistency, rigorous multi-center validation, and cross-disciplinary collaboration to support reliable clinical decision-making.

Late-stage Parkinson's disease (LSPD) represents an under-recognized and under-studied phase of PD characterized by severe axial motor symptoms, high non-motor burden, cognitive impairment, functional dependence, and substantial caregiver involvement. Despite its major clinical and socioeconomic impact, LSPD seem to be largely excluded from clinical trials, likely due to advanced disability, mobility limitations, and the lack of appropriately adapted outcome measures.To better characterize the representation of this population in research, we conducted a systematic review of clinical trials listed in major clinical trials registries, focusing on studies that included participants with a Hoehn and Yahr stage of 4 or 5. As of January 2026, 353 registered trials allowed inclusion of individuals with LSPD; however, many of these studies applied exclusion criteria related to dementia. Most trials investigated pharmacological or rehabilitation interventions targeting axial or global motor symptoms. Among the published studies (27.8% of the total), only 30% actually enrolled participants with LSPD.Our findings confirm the limited inclusion of individuals with LSPD in clinical research. Broadening eligibility criteria, implementing decentralized or home-based trial designs, incorporating caregiver-centered outcomes, and adopting cognitively adapted assessment tools may foster a more inclusive and patient-centered research framework. Such efforts are essential to better address the needs of this growing and vulnerable population.

The Banff Classification provides the global standard for evaluating renal transplant biopsies, yet its semi-quantitative nature, complex criteria, and inter-observer variability present significant challenges for computational replication. In this study, we explore the feasibility of approximating Banff lesion scores using existing deep learning models through a modular, rule-based framework. We decompose each Banff indicator-such as glomerulitis (g) , peritubular capillaritis (p t c) , and intimal arteritis (v) -into its constituent structural and inflammatory components, and assess whether current segmentation and detection tools can support their computation. Model outputs are mapped to Banff scores using heuristic rules aligned with expert guidelines, and evaluated against expert-annotated ground truths. Our findings highlight both partial successes and critical failure modes, including structural omission, hallucination, and detection ambiguity. Even when final scores match expert annotations, inconsistencies in intermediate representations often undermine interpretability. These results reveal the limitations of current AI pipelines in replicating computational expert-level grading, and emphasize the importance of modular evaluation and computational Banff grading standard in guiding future model development for transplant pathology.

Virtual reality (VR) has gained traction in psychiatric research in exposure-based interventions for phobias, simulations of hallucinations in psychosis, and adaptations of cognitive-behavioral therapy. These applications demonstrate that emotionally immersive digital environments can meaningfully engage pathological processes in controlled settings. Yet VR's potential in psychiatry goes beyond providing new contexts for well-established therapeutic interventions. Here, we argue that while emotionally rich, interactive VR scenarios can be leveraged to radically improve translational experimental design in behavioral neuroscience, there are not yet clear conceptual frameworks for the incorporation of these elements. Traditional lab tasks with theoretical implications for therapeutic interventions often rely on abstract procedures that fail to capture the immersive, affective dynamics of real-world maladaptive behaviors. Here, we discuss how high-fidelity VR scenarios simulate complex emotional naturalistic contexts while retaining precise experimental control. We illustrate this drawing on our recent work using high-fidelity, game-engine-based VR scenarios and consider how these environments offer a scalable and accessible platform for large-sample studies, as VR hardware becomes more common in the household. We propose a set of dimensions called the Immersive Behavioral Translation Framework (IBTF) that suggest conceptual considerations for VR in neuroscientific research: (i) The controllability of immersive environments; (ii) features of immersive environments that bridge ecological validity gaps; and (iii) the relatability of these emotionally immersive experiments to clinical conditions in translational research. We illustrate this framework with reference to emerging findings in associative learning and threat sensitivity, while emphasizing the broader implications for rethinking how we study cognition, emotion, and psychopathology.

Large language models (LLMs) represent a type of generative artificial intelligence (GenAI) that generate and interpret text, with some LLMs able to process multimodal content (e.g., images, audio, video), and can be deployed as part of agents to perform users' tasks. LLMs can perform natural language processing functions such as summarization, translation, and extraction giving them the potential to enhance and scale pharmacoepidemiological and real-world research by performing tasks such as literature review, data extraction, and medical writing. Despite the growing integration of GenAI tools into research workflows, their technical foundations and methodological implications remain unfamiliar to many pharmacoepidemiologists, who are often responsible for the reliability and accuracy of research that relies on these tools. This paper aims to inform pharmacoepidemiologists about the capabilities and limitations of LLMs to support responsible integration into the field of pharmacoepidemiology, providing an intuitive overview of how LLMs work, focusing on training and text generation, and reviews current and emerging applications in drug effectiveness and safety research and epidemiology. The article addresses challenges associated with LLM use in real-world evidence generation, including concerns regarding reproducibility, bias, hallucinations, plagiarism, data privacy, and the need for validation.

Chemical synthesis remains a critical bottleneck in the discovery and manufacture of functional small molecules[1-3]. While AI-assisted synthesis planning has proliferated in recent years, a detailed understanding of its failure modes has not been achieved, and models still struggle with predicting less frequent, yet strategically critical reactions, as well as hallucinated, incorrect predictions misaligned with chemists' expectations[4-12]. In this work, we analyze the failure modes of current AI models and propose RetroChimera: a frontier retrosynthesis model, built upon two newly developed components with complementary inductive biases, integrated via a novel, learning-based ensembling strategy. Through experiments across several orders of magnitude in data scale, we show RetroChimera outperforms leading baselines, demonstrating robustness outside the training data, as well as the ability to learn from very small numbers of examples per reaction class. Using both pairwise and pointwise setups, we find that organic chemists prefer predictions from RetroChimera over published reference reactions and over other AI models. Finally, we demonstrate zero-shot transfer and fine-tuning on internal datasets from two major pharmaceutical companies, showing robust generalization under distribution shift. Our work demonstrates the viability of deep learning for accurate synthesis prediction in increasingly challenging regimes.

Artificial intelligence (AI) is increasingly shaping radiology, by performing tasks ranging from image reconstruction and workflow optimization to image analysis, reporting support, and multimodal clinical decision support. This narrative review provides a structured overview of the methodological foundations and clinical evolution of AI in radiology. We first outline the historical development from early rule-based expert systems and classical feature engineering to data-driven machine learning. We then explain the core principles of neural networks, including network depth, nonlinear activation functions, loss-based optimization, and backpropagation, before discussing convolutional neural networks (CNNs) as the dominant architecture for image-based learning in radiology. Particular emphasis is placed on inductive biases such as locality, parameter sharing, and translation equivariance, as well as on encoder-decoder architectures and U-Net-based approaches for segmentation. Subsequently, we discuss attention mechanisms, Transformer architectures, and their role in modeling long-range dependencies in medical imaging. We distinguish Vision Transformers, which process images as token sequences, from vision-language models, which align imaging data with textual information and enable multimodal tasks such as report generation, image-text retrieval, and visual question answering. Recent radiology-specific foundation models and large language models are considered in the context of workflow support, structured reporting, knowledge integration, and emerging multimodal image interpretation. Finally, we address key challenges for clinical translation, including data requirements, bias, generalizability, hallucinations, interpretability, validation, federated learning, and regulatory aspects such as the EU AI Act and the Medical Device Regulation. A basic understanding of these methodological principles is essential for radiologists to critically evaluate current AI systems and to contribute to their safe, transparent, and clinically meaningful implementation.