Tiny particles bounce light around in a unique way, a property that researchers are using to detect pollutants in water and soil samples.
Get the latest international news and world events from around the world.
Oral Corticosteroid Use During Pregnancy and Gestational Diabetes Risk
Among pregnant women, oral corticosteroid use was not linked to higher rates of gestational diabetes.
No substantial increase in risk was observed among women exposed to OCSs between 1 and 27 weeks’ gestation, except for a slight increase during weeks 4–6. The findings remained consistent across age groups, indications, doses, and durations.
Question Does oral corticosteroid use during pregnancy increase the risk of gestational diabetes?
Findings In this cohort study of 1 325 940 pregnancies resulting in live births, oral corticosteroid use between 1 and 27 weeks’ gestation was not associated with an increased risk of gestational diabetes compared with unexposed pregnancies.
Meaning These results suggest that oral corticosteroid treatment during pregnancy does not elevate the risk of gestational diabetes and may be considered a reasonable therapeutic option for managing maternal conditions.
Controllable deterministic quantum teleportation of multiple sideband qumodes
As an important quantum communication protocol, quantum teleportation has broad applications in quantum information science and technology. For the application, it is essential to enhance the ability of quantum teleportation by teleporting multiple quantum states simultaneously. Here, we experimentally demonstrate deterministic quantum teleportation of multiple sideband qumodes of coherent states, which corresponds to teleporting multiple quantum states simultaneously, with the assistance of continuous-variable (CV) quantum entanglement. By fine-tuning the phases of two classical channels according to different adjustable frequencies, we successfully realize deterministic CV quantum teleportation with up to 5 sideband qumodes simultaneously within the frequency bandwidth of 24 MHz.
Stem cell patch reverses brain damage in fetuses with spina bifida
A patch made of stem cells from donor placentas has been used to treat fetuses in the womb with a severe form of spina bifida as part of a world-first trial. The novel approach seems to have reversed a brain complication associated with the congenital condition at least as effectively as the go-to treatment, but is expected to enable more children to walk over the long term.
The mother of one of the babies, who is now 4 years old, says she expected that her son Toby would require a wheelchair when he was diagnosed with the condition in the womb. “But Toby is healthy [and] has hit all of his milestones – he’s walking, running and jumping – and has no problems with bladder control, which is rare for people with the condition,” she says.
Spina bifida – which affects about 1 in every 2,800 births in the US every year – occurs when a baby’s spine and spinal cord do not fully develop in the womb. In the most severe form of the condition, called myelomeningocele, the spinal cord and its surrounding tissue protrude out of a gap in the vertebrae, which often impairs mobility and bowel and bladder control. The cause of spina bifida is unknown, but folic acid deficiency during pregnancy raises the risk.
One of the standard treatments involves surgery in the womb that tucks the spinal cord and the surrounding tissue back into the vertebrae, before sewing up the skin to form a tight seal. “But many children still end up unable to walk and there’s [usually] no improvement in bowel or bladder control,” says Diana Farmer at the University of California, Davis.
This led Farmer and her colleagues to wonder if the addition of stem cells could help by promoting the growth and repair of spinal tissue. To find out, they recruited six pregnant women carrying fetuses with myelomeningocele.
Read More
DeepSeek withholds latest AI model from US chipmakers including Nvidia, sources say
SAN FRANCISCO/SINGAPORE — DeepSeek, the Chinese artificial intelligence lab whose low-cost model rattled global markets last year, has not shown US chipmakers its upcoming flagship model for performance optimization, two sources familiar with the matter said, breaking from standard industry practice ahead of a major model update.
Instead, the lab, which is expected to launch its next major update, V4, granted early access to domestic suppliers, including Huawei Technologies, the sources said.
AI developers typically share pre-release versions of major models with leading chipmakers such as Nvidia and Advanced Micro Devices to ensure their software performs efficiently on widely used hardware. DeepSeek has previously worked closely with Nvidia’s technical staff.
A Developmentally Informed Study of Sleep and Circadian Polygenic Scores in Adolescence
Adult sleep GWAS-derived polygenic scores demonstrated comparable associations with corresponding sleep phenotypes in Adolescents, suggesting genetic influences on sleep persist across developmental stages.
Question Do genetic variants that are associated with adult sleep/circadian phenotypes influence sleep phenotypes in adolescents?
Findings In a population-based birth cohort study (N = 3903), genetic influences on all adult sleep phenotypes (sleep duration, insomnia, daytime sleepiness, napping, and chronotype as indexed by polygenic scores derived from adult genome-wide association studies) were associated with their corresponding sleep/circadian phenotypes in adolescents aged 15 years.
Meaning Genetic variants identified in adult genome-wide association studies may also be relevant to a variety of sleep phenotypes in adolescence, suggesting that these variants index sleep phenotypes during a key developmental stage in which sleep disturbances typically emerge.
Foundation Models Meet Medical Image Interpretation
In contrast, traditional deep learning methods in the medical domain have long been constrained by scarce annotations data, weak cross-modal semantic correlation, and insufficient generalization capabilities. FMs can effectively alleviate these issues by extracting semantic representations from large-scale unlabeled data, reducing dependence on expert annotations, and enhancing cross-modal understanding and transferability [7]. This provides technical support to address challenges such as long-tail distributions, data scarcity, and modality imbalance, thereby promoting a shift in medical decision-making from experience-driven to data-driven approaches.
Unlike traditional specialist models such as nnU-Net [8], which are typically designed for a single modality and specific tasks, FMs emphasize modality unification and task generalization, enabling cross-domain transfer and knowledge sharing. With mechanisms such as prompt engineering and PEFT, these models support few-shot and even zero-shot transfer (ZST). For example, Med-PaLM [9] is based on a unified medical pretraining model, which can generate structured pathology reports and perform lesion localization from medical images. It effectively overcomes the limitations of traditional methods that require separate architectures for different tasks, significantly improving modeling efficiency and system integration. Driven by such unified model architecture, medical AI systems are evolving toward greater generality and reusability.
Despite these advancements, the unique characteristics of the medical domain pose multiple challenges to the application of FMs. On one hand, medical data are highly heterogeneous, with pronounced differences in resolution, contrast, and noise distribution across imaging modalities such as computed tomography (CT), magnetic resonance imaging (MRI), and ultrasound [10]. This limits the ability of traditional single-modality pretraining strategies to achieve effective cross-domain knowledge integration. On the other hand, clinical applications demand higher standards for model performance. Clinical decision-making relies on interpretable diagnostic evidence, yet pretraining models often behave as “black boxes”, limiting their clinical traceability [11]. In addition, the long-tail distribution of rare diseases poses fairness challenges for model generalization [12].