Section
18 articles
Melanoma is hard to treat because it grows, spreads and learns to ignore chemotherapy. A team in China built a dissolving microneedle patch for mice: tiny needles carrying doxorubicin and a CRISPR/Cas9 complex aimed at the YB-1 gene. The question was whether pressing such a patch onto skin could deliver both agents close enough to the tumor to change its behavior.
Blood pressure is usually checked with a cuff that squeezes the arm and gives a single number. But the pulse itself carries more information: its shape changes as it travels through arteries, and those changes can hint at how blood moves in the body. A new passive wrist patch tries to turn that hidden detail into something an ordinary phone camera can see. The device has no battery or wires. It uses a stretch-sensitive film whose color shifts when the artery beneath it beats.
Older adults leave long trails of medical codes in insurance records: diagnoses, procedures, prescriptions, and the months between them. Those trails can hide patterns that are hard to see by eye, especially when a treatment might be linked to a later decline in memory or thinking. The question is whether routine claims data can be turned into a practical way to spot such patterns before they become testable drug ideas.
For a blind person moving through a room, the hardest part may be knowing where things are before reaching out. One answer is to stop describing objects in words and instead give them a place in space as sound. A wearable pair of glasses can detect nearby items and play short everyday sounds around the head, so a bottle might feel like it is coming from the left while a cup sits ahead. The question is whether that kind of spatial audio can guide hands reliably enough to matter.
In many homes, a child's bedtime is treated as a routine to manage rather than a sign to read. For preschoolers with autism, that routine can look different from the start: nights may begin later, end later, and be broken more often. The question is whether those patterns are visible early enough to matter.
During radiotherapy, a quick cone-beam scan helps aim treatment at the tumour. The same image is often too grainy and distorted to calculate dose safely, so clinicians may order another CT scan. That extra scan adds radiation, cost, and waiting time. A new way of rebuilding a cleaner, CT-like volume from the noisy scan could change that workflow by using the scanner's own geometry as a built-in check.
In Portugal's emergency departments, a hospital's forecast should fit inside its region, and every region should fit inside the country. Yet many demand models treat each level as separate, leaving planners with numbers that look precise but do not reconcile. The question is whether one system can predict patient visits 28 days ahead while keeping hospital, regional and national totals in line.
A protein linked to Parkinson’s disease can build up in specific brain regions long before movement problems appear. Scientists used a rat model that carries the human gene for this protein and tested whether reducing it early could change behavior, signaling, and regional brain chemistry.
High-frequency ultrasound can image the thin layers of skin, from epidermis to muscle, but automated analysis is held back by a shortage of expert labels for deeper structures. A physics-based simulator offers one route around that bottleneck: it builds virtual skin phantoms and generates paired scans with exact layer masks. The open question is whether those synthetic images contain enough realistic structure and texture to prepare a model before it sees real patient data.
Hospitals keep detailed records of diagnoses, procedures, lab tests and medicines for every admission. Those structured entries are usually treated as dry administrative data, while the richer story lives in doctors' notes. But predicting whether a patient will come back within a month may not require reading any prose at all. Turning routine hospital codes into named medical concepts makes it possible to estimate readmission risk from the record itself.
A short recording of your breathing can carry more than rate and depth. Sixty-second breath traces from a public dataset, recorded while people sat at baseline, felt stress, amused themselves, or meditated, are checked for subtle timing, rhythm, and waveform patterns that might separate those states without asking anyone how they felt.
When someone leaves an emergency department and returns within two weeks, clinicians may wonder whether the first visit missed something important. Hospitals often review only a narrow slice of these returns because full chart reviews are slow and expensive. Doctors and artificial intelligence were asked to judge which return visits deserve closer scrutiny, using only brief diagnosis labels.
A contact-tracing app that warns users of infection risk also creates a narrow path for an attacker to learn whether one specific person is sick. The question is whether the warning can be made private enough to protect individuals while still stopping outbreaks, without turning every alert into useless noise.
Choosing a blood test, scan, or lab value is often a judgment call. A patient may have several possible explanations for the same symptoms, and each test can point in a different direction. New work asks whether a language model can help by imagining what each result might show before the sample is drawn.
Brain metastases can range from large, obvious masses to specks barely visible in a 3D MRI scan. A computer system must find every one of them while avoiding false alarms that could send patients toward unnecessary treatment. One way to test whether scale matters is to give the same type of computer model two different views of the brain: a wide view with more context and a tighter view where small lesions occupy more space.
A promising molecule in a lab flask must pass through years of human testing before it can reach a pharmacy shelf. Along the way, trial designers face a stubborn problem: how to learn whether a treatment is safe and useful without wasting time, money, or patients on approaches that will not work. The path from discovery to approval now includes adaptive designs, digital sensors, gene therapies, AI tools, and regulators who are rewriting rules faster than ever.
Electronic health records contain a long stream of measurements, notes, scans, and decisions made during hospital care. A machine-learning system trained on that stream treats each patient stay as one continuous sequence and asks what kind of event may come next, how much time will pass before it arrives, and whether changing an earlier treatment choice would alter the later course.
Hospitals are turning to software robots for billing, scheduling, insurance checks, and records work. The hard part is not buying bots; it is deciding which administrative tasks should be automated first, what tools can safely handle protected health information, and how much each project will cost or save.
