Search the evidence

Find the signal.

Search titles, impact clusters, countries, organizations and the full text of every analysis.

5 stories found

A clinician studies a generic brain scan on a monitor in a softly lit clinical room.
Social good & healthClinical research+1 clusters01

AI spots tiny brain bleeds. The next test is the clinic.

A useful AI result does not need to promise the end of doctors. A newly published Scientific Reports study evaluates software designed to detect cerebral microbleeds on MRI. Its appeal is practical: finding small abnormalities consistently is a task worth improving. In the retrospective, single-centre test, the standalone system achieved 88.35% lesion-level sensitivity with 0.95 false-positive detections per patient. It met the study’s prespecified performance targets. Those are encouraging measurements, not proof that the software improves patient outcomes or reduces a clinician’s workload in routine use. The unit matters. Detecting most individual lesions is different from identifying whether a person has any lesion at all. Neither number is an all-purpose “accuracy” score. Extra findings can create review work; missed findings remain relevant. The test did not establish that replacing a reader produces better care. The commercial context also deserves daylight. Neurophet sponsored the study, and several coauthors were current or former employees who may hold shares. That does not invalidate the results. It strengthens the case for independent replication and evaluation across different hospitals and scanners. This is a good example of progress that can be welcomed without overselling it. The next valuable evidence would connect detection to decisions: reader performance, time spent resolving false alarms and whether the tool helps in a broader clinical population. A bounded success is a stronger foundation for adoption than a sweeping replacement claim.

3 min
Cognition & learningGlobal+2 clusters02

Souei et al., “Artificial intelligence in deep brain stimulation for movement disorders: a systematic review and technology readiness assessment”

Researchers reviewed 239 peer-reviewed studies on AI-supported deep-brain stimulation and found a pronounced gap between reported algorithmic performance and clinical readiness. External validation remained rare, evaluations were predominantly retrospective and single-centre, and more than one-quarter of studies used small, high-dimensional datasets with elevated overfitting risk; most systems therefore remained at early-to-intermediate technology-readiness levels.

2 min
Layers of translucent teal and blue paper form a brain above a slide speckled with violet marks. This is a concept illustration, not a patient scan.
Social good & healthInternational research cohorts+1 clusters03

Which children could safely get less radiation? A tumor test offers a clue

A family dealing with a childhood brain tumor faces two different futures at once. Parents want the treatment most likely to keep the cancer from coming back. They also want to protect the child who will have to live with what that treatment does to memory and learning. A new Nature Genetics study offers a way to ask more precisely which children might benefit from less radiation. The researchers built MANTIS using tumor DNA methylation profiles from 2,786 patients across two diseases, medulloblastoma and ependymoma. Methylation marks are chemical tags on DNA; many centers already use the patterns to help classify brain tumors. MANTIS uses those data to estimate an individual survival curve over ten years. The medulloblastoma model was validated in three separate clinical-trial cohorts totaling 898 patients. That is a test against real follow-up data, not a trial in which a model chose a child’s treatment. The team then revisited radiation doses for average-risk medulloblastoma. Its analysis singled out roughly 60% of the patients studied in that setting as possible candidates for lower-dose radiation to the brain and spine. The researchers did not detect a statistically significant survival disadvantage in that selected group; lower doses were associated with better cognitive outcomes. A difference too small to detect in one analysis can still matter enormously to a child. The promising result is a testable way to select patients for a trial, not evidence that six in ten children receiving cancer care should have treatment reduced.

4 min
A young person at a window has both a phone and possible human connections beyond it.
Social good & healthGlobal+2 clusters04

Some teenagers find AI support helpful. Many still want more help.

If a young person says a conversation helped, I think we should listen. We should also ask the next question: did they get the support they still needed afterwards? A new peer-reviewed study of 24,132 students in England offers an unusually useful distinction. Past-year use of general-purpose AI for mental-health support was reported by 662 students, or 2.7%. Among those answering the helpfulness question, 324 of 580—55.9%—called it helpful. But 410 of 649 respondents to a different question, or 63.2%, wanted additional support. The denominators differ and should not be merged. Most AI users, 610 of 662, also accessed other support. That makes a simple replacement story unconvincing. The chatbot may be one place a person turns within a larger, incomplete support network. The study does not establish that AI caused their difficulties, improved a clinical condition or displaced human care. There is room here for both empathy and caution. Dismissing every helpful interaction would dismiss the experience being reported. Selling that experience as proof of treatment effectiveness would ask it to carry evidence it does not contain. A useful institutional response would investigate who cannot obtain sufficient help, how young people move between digital and human support, and whether referrals actually lead somewhere. This accepted article was published on October 27, but the accessible abstract does not establish the survey field dates. Its percentage should not be presented as a current national adoption estimate.

3 min
A protected neural signal travels through an AI infrastructure pipeline toward healthcare, research, and consequential decision gates.
PrivacyEuropean Union+3 clusters05

European advisers want neuro-AI governed as infrastructure

Europe's ethics advisers are asking policymakers to stop treating neuro-AI as a collection of futuristic devices. Their new statement defines neuro-AI infrastructures as interconnected systems through which neural data is collected, processed, reused, and turned into AI-powered applications. That shift matters because the most consequential output may not be the original brain signal. It may be a derived inference about attention, emotion, health, capacity, or intent that is generated later, combined with other data, and used in a different context. The European Group on Ethics recommends stronger protection for both neurodata and neurodata-derived inferences, safeguards against disproportionate control in consequential settings, responsible development of brain foundation models, more public-interest governance capacity, and a targeted review of the existing EU legal framework. The opportunities are substantial in healthcare, rehabilitation, and research. So are the institutional risks. A consent form tied to one headset or clinical encounter may not govern an expanding pipeline of models, vendors, secondary users, and future inferences. An infrastructure approach asks who controls the data layer, which uses remain prohibited, whether people can contest derived claims, and whether Europe retains public capacity rather than relying entirely on private platforms. The statement is advisory, not law, and does not resolve which neural inferences are reliable. Privacy rules built around collection can fail when value and harm emerge through recombination. Governance must follow the signal through the whole system.

5 min