With artificial intelligence making deep inroads into healthcare, two top doctors are urging physicians to embrace the technology but to remember that only humans can serve the best interest of individual patients.
The University of Virginia School of Medicine's Andrew S. Parsons, MD, MPH, and Harvard Medical School's Adam Rodman, MD, MPH, note in a new commentary that AI is growing increasingly sophisticated at an astonishing pace. The technology can already match or exceed human doctors at certain tasks, such as making diagnoses from written cases and electronic health records. That has prompted debate about whether AI will soon replace doctors for many applications.
Parsons and Rodman, however, note that deciding how to manage a patient's care -- choosing treatments, tests and follow-up that fit the patient's goals and circumstances - cannot be reduced to a "correct" answer. AI may be able to identify a patient's medical condition, but deciding what to do next means choosing which risks a particular patient should accept, and a recommendation that works in theory can fail in practice.
"Guidelines and trials skew toward the populations studied, and real patients frequently fall outside those norms because of complex comorbid conditions and social circumstance or because their goals differ from a trial's end points," Parsons and Rodman write. "Deviating from the technically correct answer is often good medicine, and someone must answer for that deviation."
And there lies the real role for human doctors in the medicine of tomorrow, the authors argue. Even as AI takes on more of diagnosis, doctors must consider each individual's circumstance, work with patients to make hard choices, then shoulder the responsibility for the outcomes.
"We contend that the clinician's enduring role is accountability: owning the risk of acting under uncertainty and answering for the decision that follows," Parsons and Rodman note. "That ownership is real only when the clinician still reasons about the decision and holds a stake in its outcome."
Kaohsiung Medical University Chung-Ho Memorial Hospital has been working with industry and hospital partners worldwide to develop and implement new AI tools, shares Deputy Superintendent and Professor Horng-Shing Lu.
MedStar Health Research Institute and Regenstrief Institute will develop tools to support patient-controlled AI workflows and improve laboratory data exchange.
A group of scientists led by Joseph Brown, assistant professor at the Leslie Dan Faculty of Pharmacy, has received $2.3 million USD in funding to develop a database and AI training model to identify peptides with therapeutic potential. The project is funded through the Google.org Impact Challenge: AI for Science initiative.
The newly funded initiative, OpenPep, will lead to an open access database that will allow researchers to use artificial intelligence (AI) to predict which peptides will bind to target proteins.
The HIMSS AI in Healthcare Forum in San Diego will explore all angles of using artificial intelligence to advance health, says Dr. Amy Sitapati, executive director of the Institute for Applied Health Intelligence at UC San Diego Health.
A team from the Universitat Politècnica de València (UPV) and the University of Valencia (UV) has developed and validated an artificial intelligence (AI) system capable of distinguishing Ewing sarcoma from other morphologically similar tumors with high accuracy. The model, which also enables the full optimization of tissue use in biopsies, achieves 91.6% accuracy.
This breakthrough, published in the scientific journal International Journal of Molecular Sciences, represents a qualitative leap forward in digital pathology by aiding the diagnosis of very complex pediatric and juvenile neoplasms; it reduces errors and preserves the scarce biological material obtained during needle biopsies.
About Ewing sarcoma
Ewing sarcoma is a rare malignant tumor that primarily affects children, adolescents and young adults. It usually occurs in the bone, although it can also develop in soft tissues. One of the main difficulties in diagnosing it is that, under the microscope, it can be very similar to other tumors that also originate in soft tissues, such as rhabdomyosarcoma (striated skeletal muscle), chondrosarcoma (cartilage), gastrointestinal stromal tumors (the digestive tract) - GIST, as it is known - or synovial sarcoma (joints).
Researchers from the Department of Pathology at the University of Valencia - Francisco Giner (lead researcher and first author of the article), Isidro Machado and Samuel Navarro (who is also coordinator of the Paediatric Solid tumor Translational Research Group at the INCLIVA Institute for Health Research) - explain that "in many small biopsies, there is very little tissue. If the pathologist has an initial tool that accurately guides the suspected diagnosis, they can decide which additional and immunohistochemical tests to request, thereby avoiding using up the sample on analyses that do not provide decisive information".
AI to assist medical staff
The development of the new AI tool has been led by a team from the CVBLab research group at the Universitat Politècnica de València (UPV) and the company Artikode Intelligence, a spin-off from the UPV.
Nanyang Technological University, Singapore (NTU Singapore) and the Institute of Mental Health (IMH) have developed a novel artificial intelligence (AI) framework capable of identifying youths at high risk of psychosis - a severe mental health condition that typically emerges in adolescence to early adulthood where a person may experience hallucinations, delusions, and disordered thinking - from short samples of casual, spontaneous speech.
This is the first reported study in Singapore to focus specifically on early psychosis risk among youths using large language models (LLMs), a form of artificial intelligence trained to analyse and generate human language. The team of researchers believes the technology could eventually support primary care providers and community workers in identifying vulnerable youths before clinical symptoms escalate.
Early identification of individuals at ultra-high risk (UHR) of psychosis is challenging because youths in everyday life rarely articulate mental health struggles directly. Ultra-high risk of psychosis refers to a clinical state in which an individual shows early warning signs or risk factors that significantly increase the likelihood of developing psychosis.
For example, a depressed person may not explicitly state that they feel depressed. This is known as "non-disclosing speech," meaning language that does not contain any explicit mention of the mental health condition experienced by the speaker. Therefore, evaluating whether AI can detect potential risk purely from casual, spontaneous conversation addresses a key hurdle in early intervention.
Transforming casual conversation into clinical insight
Published in the scientific journal npj Mental Health Research, the joint NTU-IMH study evaluated transcribed text samples from 429 participants aged 14 to 29 enrolled in Singapore's Longitudinal Youth-at-Risk Study (LYRIKS), a landmark Singapore study initiated in 2008 to comprehensively assess the social, clinical, and biological risk factors associated with youths at ultra-high risk of developing psychosis.
Each participant provided a 45-second non-self-disclosing speech sample in response to a neutral prompt, such as making a case for why they should win an award for being the 'most interesting person'.
The researchers found that standard LLMs struggled when prompted directly to predict clinical risk from raw transcripts, achieving performance no better than a random guess and missing most at-risk individuals.
The researchers believe standard models struggled due to a lack of suitable training datasets, as well as the nature of spontaneous everyday speech, which rarely contains structured clinical information or explicit symptom descriptions.
To overcome this limitation, the team developed LEMon-AID (Latent Emotion Proxy for Mental Health Detection with AI-integrated Decision-making), which leverages LLMs to extract subtle emotional profiles such as joy, trust, fear, and sadness from the text, then combines these emotional signals with traditional machine learning predictions.
The study demonstrated that LEMon-AID performs significantly better than conventional approaches. It boosted overall predictive accuracy from 55 per cent under conventional direct LLM prompting to 67 per cent, while dramatically increasing detection sensitivity such that up to 93 per cent of at-risk youths can be identified.
Human-AI collaboration and transparent reasoning
Co-lead investigator of the study, Wilson Goh, Assistant Professor of Biomedical Informatics from LKCMedicine and the NTU Centre of AI in Medicine, said: "Standard AI models struggle to pick up early mental health risks from everyday conversation because people rarely talk about their symptoms openly. Instead of asking AI to give a direct diagnosis, LEMon-AID uses language models to read underlying emotional cues. This system connects emotional patterns with mental health risks while keeping patient data private and providing clear, understandable explanations."
Unlike traditional machine learning models, which may operate as 'black boxes', LEMon-AID generates natural-language explanations alongside its risk assessments. This allows clinicians to understand the clear reasoning behind each prediction, making AI a safer and more transparent tool for doctors to use when evaluating patients.
Dr Jimmy Lee, Group Chief Research and Innovation Officer, NHG Health and Senior Consultant, IMH, and an Associate Professor at NTU's LKCMedicine, who led the LYRIKS study said, "Early detection is one of our most powerful tools in preventing long-term disability and restoring the health and futures of young people with mental illness. This research brings us a meaningful step closer to that goal, a future where frontline community workers and healthcare providers are better equipped to identify at-risk youths early, supported by AI tools that help surface those who might otherwise be missed. The aim is to reach young people sooner and encourage them to seek the care they need before their condition worsens."
The guidance addresses emotional dependence, AI-based self-diagnosis and privacy risks following a national survey that found widespread use of generative AI for emotional support.
The first Europe-wide expert working group on Artificial Intelligence in mental health launches today to the ECNP congress in Munich, where the group will host an open discussion, setting out the potential, the problems, and possible directions for AI in psychiatry and neuroscience. Going forward, the group will set out plans to guide the regulation and development of artificial intelligence (AI) in mental health in Europe.
Group Chair, Dr Paris Alexandros Lalousis (of King's College, London) said.
"AI is developing at a rapid pace. It offers great opportunities in mental health, but the development of AI also needs to be guided, communicated, and regulated, and this is something that is missing at the moment. The group will set out a framework for understanding the risks and benefits of applying AI in mental health, with recommendations on how we take the field forward. We are the first group to undertake such a challenge. We will map the field, so that Europe-wide resources are available for researchers to understand who is doing what, who is regulating AI, and so on. The aim is to produce a level-headed, expert-driven framework which will deal with the major issues".
The group is being established by the ECNP, which is Europe's largest clinical neuroscience organization, with interests stretching from psychiatry through to drug development, psychology, and neuroscience.
Co-Chair Tim Hahn (from the University of Muenster, Germany) added:
"Our working group comprises 19 experts, from Europe and the US. We have already developed an initial work plan for the next 3 years, and we are now meeting with the wider psychiatric and neuroscience community to feedback and refine our plans.
We are aiming for a coordinated AI framework at a European level. There are many talented people across Europe working in the field, but often they work in silos, and we need to enable communication and interaction. So I think my message would be that whether you see AI through a positive lens or a negative lens, we would like to have you engage with us".
The working group currently has 19 members from Europe and the US, who undertake research in a variety of fields, for example using basic machine learning to do drug discovery, patient concerns, basic neuroscience, through to building and evaluating chatbots.
Tim Hahn continued, "We have a great variety of expertise, but even so the biggest challenge is going to be to really understand what people are doing right now and where should we insert ourselves as a group, as a network, to provide guidance. We want to establish a structure for this huge amount of information".
The Beehive State is establishing a verification mechanism for its healthcare artificial intelligence pilot sandbox program. A new network of third-party evaluators aims to ensure that developers' AI performance and safety claims can be trusted.
The retrospective, two-institution study used 13,633 patient records to assess whether fine-tuned large language models (LLMs) could accurately differentiate epilepsy from PNES.
EpiScreen outperformed the comparison models, achieving an area under the receiver operating characteristic curve (AUC) of up to 0.992 in within-institution testing and 0.980 in cross-institutional testing. AUC measures a model’s ability to distinguish between the two conditions across different classification thresholds.
In a retrospective case-review assessment, EpiScreen improved clinicians’ classification accuracy, supporting its potential in future epilepsy screening applications.
Background
Epilepsy is clinically defined as a chronic neurological disorder characterized by recurring, unprovoked seizures. Seizures arise from sudden, abnormal electrical discharges in the brain, with public health records estimating that approximately 50 million individuals live with the condition globally.
Epilepsy’s diagnosis can be complicated by PNES, also termed functional or dissociative seizures. These involve paroxysmal events (sudden, temporary seizure-like episodes) that resemble epileptic seizures but are not caused by epileptic electrical activity.
Since epilepsy and PNES have different underlying mechanisms and management needs, timely differentiation and accurate diagnoses are critical for appropriate care. Misdiagnosis remains common, with prior studies indicating that 20%–30% of patients referred to tertiary centers for presumed drug-resistant epilepsy receive a PNES diagnosis.
Drinking water sources can contain diverse bacterial and viral pathogens, making reliable monitoring essential for public health protection. Conventional surveillance commonly relies on fecal indicator bacteria, but these organisms do not consistently reflect the presence of specific pathogens, especially viruses. Culture methods can take several days, while molecular testing requires specialized equipment and trained personnel. Machine learning offers a way to estimate microbial concentrations from routinely collected environmental measurements, capturing relationships that simple linear models may miss. However, much previous work has focused on indicator bacteria rather than pathogens themselves. Connecting pathogen predictions with health risk assessment therefore remains an important research need, alongside determining whether these models can provide useful, interpretable information across changing environmental conditions and different monitoring periods.
A study (DOI: 10.48130/biocontam-0026-0009) published in Biocontaminant on 10 July 2026 by Changzheng Cui's team, East China University of Science and Technology, shows that environmental measurements could support predictions of multiple microbial targets and generate potential health risk estimates broadly consistent with calculations based on measured pathogen concentrations.
The researchers collected 95 surface water samples from five fixed sites across two drinking water sources in an eastern Chinese city between May 2024 and December 2025. They measured seven water quality parameters, including temperature, turbidity, conductivity, dissolved oxygen, and total dissolved solids, and incorporated six meteorological variables. Membrane filtration quantified three indicator bacteria, while quantitative polymerase chain reaction measured six pathogens: Pseudomonas aeruginosa, Salmonella spp., Shigella spp., adenovirus, norovirus, and enterovirus. Six algorithms were compared, spanning linear regression, boosting, decision trees, support vector machines, random forests, and neural networks. Sampling batches were kept together when dividing training and test datasets, and grouped cross-validation supported model optimization.
Additional samples collected in January and February 2026 provided independent temporal validation. Microbial measurements revealed frequent contamination and weak or inconsistent relationships between bacterial indicators and viruses. Random forests and decision trees performed best for most targets. The selected models achieved reported coefficients of determination above 0.75, with the decision tree for P. aeruginosa exceeding 0.90. Temporal validation generally supported predictive performance, although uncertainty remained greater for several pathogens. SHapley Additive exPlanations analysis identified turbidity as a dominant contributor to indicator bacteria predictions, while pathogen models depended on differing combinations of environmental variables. The team then incorporated predicted concentrations into quantitative microbial risk assessment and used 10,000 Monte Carlo iterations to characterize uncertainty. Predicted disease burden distributions broadly matched those calculated from measured concentrations. Under the modeled exposure and treatment assumptions, Salmonella, Shigella, and enterovirus had estimated probabilities of 31.68%, 59.70%, and 20.86%, respectively, of exceeding the study's health burden benchmark. Disinfection efficiency was the strongest driver of risk variability. These estimates describe potential risks under specified scenarios; molecular detection measures genetic material and does not independently establish pathogen infectivity or actual illness among consumers. The findings also highlight treatment performance as central to controlling microbial exposure.
The study provides a practical route from routine environmental monitoring to interpretable estimates of microbial hazards and potential health consequences. Further validation across independent watersheds, seasons, and treatment settings is needed before wider deployment. Integration with real-time monitoring platforms could eventually support targeted testing, timely interventions, and more responsive management of drinking water sources.
