Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain
the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in
Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles
and JavaScript.
A vision–language foundation model, trained on a dataset of more than 1 million echocardiogram video–text pairs, is able to assess various cardiac structural and functional parameters despite not having been directly trained on any specific image interpretation task.
In a randomized clinical trial, alerts based on the detection of abnormalities in electrocardiograms using a deep learning algorithm reduced all-cause mortality at 90 days in patients admitted to hospital emergency or internal medicine departments.
In a case series of six patients with multidrug-resistant rheumatoid arthritis, the CD19xCD3-targeting bispecific T cell engager blinatumomab reduced disease activity and led to reductions in autoantibodies.
Tailored to detect and prevent potential medication direction errors in a digital pharmacy data processing pipeline, a large language model is shown to increase efficiency and decrease burden for technicians and pharmacists in a prospective application.
In a phase 2 trial, nivolumab achieved a response rate of 58% in patients with mismatch-repair-deficient gynecological cancers, meeting the primary endpoint, and genomic and immunologic features correlated with response.
An antibody screen of two distinct multiple sclerosis cohorts reveals an autoantibody signature that is detectable years before symptom onset and linked to a common microbial motif.
In a series of clinically relevant tasks in computational pathology, AI-driven models display marked performance disparities across demographic groups, which can only partially be mitigated by self-supervision on large training datasets and existing debiasing techniques.
Implementation of organized low-dose computed tomography screening in over 4,000 individuals with high risk for lung cancer as part of the Ontario Lung Cancer Screening Pilot reported high cancer detection rates, early detection of cancer and low serious harms.
The QR4 algorithm for prediction of 10-year cardiovascular disease risk, developed, tested and externally validated in datasets comprising 16.8 million people from the United Kingdom, improves upon the QRISK3 algorithm that is in current use by incorporating new risk factors.
A self-amplifying mRNA vaccine shows promise in this new modality by eliciting neutralizing antibodies against the SARS-CoV-2 Omicron (BA.1) variant in a phase 2/3 trial.
Using plasma samples collected over several time points during pregnancy from three different cohorts, associations between circulating placental IGFBP1 levels, metabolic traits and birth anthropometric measurements were measured, with low IGFBP1 levels identified as a potential risk factor for gestational diabetes mellitus.
Developed on cytology images of hydrothorax and ascites from 57,220 cases at four hospitals, a deep-learning model shows high accuracy in tumor origin prediction and presents prognostic value when patient treatment is consistent with the cancer origin predicted by the model.
By learning to pair dermatological images and related concepts in a self-supervised manner, a visual-language foundation model is shown to have comparable performance to supervised models for concept annotation and is used to scrutinize model decisions for enhanced interpretability and accountability of medical imaging applications.
An exploratory analysis of the 1-year clinical trial PASADENA in individuals with early-stage Parkinson’s disease suggests that prasinezumab might reduce motor signs progression to a greater extent in those with more rapidly progressing disease.
In an observational study evaluating functional precision medicine in children and adolescents with relapsed or refractory solid and hematologic malignancies, it was feasible to provide personalized treatment recommendations to treating physicians on the basis of genomic profiling and ex vivo drug sensitivity testing within 4 weeks.
By generating synthetic image samples specific to underrepresented groups, diffusion models help medical image classifiers to achieve greater fairness metrics across a variety of medical disciplines and demographic attributes.
In a large multinational cohort study, maternal, gestational or pregestational diabetes was associated with only a small-to-moderate risk of ADHD in offspring, contrary to previous estimates that showed stronger effect sizes, attributing the differences in findings to confounding by shared genetic and familial factors.
The analysis of continuous glucose monitoring measurements from a large cohort of nondiabetic individuals uncovered large inter- and intraindividual variabilities, with potential implications for current diagnostic cutoffs for diabetes diagnosis and several cardiometabolic clinical measures.