Neurosurgery ·
Fine-tuned language model predicts stroke outcome from admission notes; meta-analysis finds 2D angiography misses half of clipped aneurysm remnants

Listen to this post
read by AI voices · all episodesTHE SHORT READ
A large language model fine-tuned on admission notes predicted 90-day functional outcome after acute ischemic stroke with an area under the curve of 0.87, according to a retrospective study in Neurosurgery.
A meta-analysis of 38 studies in the Journal of Neurosurgery found remnants in 23.5% of clipped intracranial aneurysms on 3D angiography and a 51.7% false-negative rate for 2D angiography.
THE DETAILS
- AI Nov 19, 2026A retrospective study in Neurosurgery fine-tuned large language models on admission notes from the NYU Langone GWTG-Stroke registry to predict functional outcome after acute ischemic stroke. A fine-tuned Llama model reached an area under the receiver operating characteristic curve (AUROC) of 0.84 on 9,485 notes for discharge outcome and 0.87 on 1,898 notes for 90-day outcome. The authors said fine-tuned models outperformed the prognostic baselines they evaluated; the study measured discrimination on past records, not effects on care.Sources: Neurosurgery, PubMed
- Research Nov 20, 2026A systematic review and meta-analysis of 38 studies in the Journal of Neurosurgery estimated that 23.5% (95% CI 19% to 28.7%) of clipped intracranial aneurysms had remnants on 3D digital subtraction angiography, and that 2D angiography had a false-negative rate of 51.7% (95% CI 38.6% to 64.6%) for detecting them. Recurrence was 21.5% after incomplete clipping versus 0.7% after complete clipping (relative risk 15.7, 95% CI 7.5 to 32.7). The authors said one in four to five remnants regrows.Sources: Journal of Neurosurgery, PubMed
- Technology Nov 19, 2026Spryte Medical said Nov. 19 that nOCT imaging was completed in all subjects in INSYTE, a prospective, single-arm, multicenter pivotal trial of its investigational intravascular optical imaging device in the neurovasculature, producing diagnostic-quality images longer than 50 mm against a predefined performance target of 24 mm. Imaging was associated with a change in investigator treatment decisions in more than 40% of patients, the company said, but the trial was not designed to test whether those changes improved clinical outcomes. No serious adverse events were directly attributed to the imaging, according to the release.Source: Spryte Medical (GlobeNewswire)
The weekly innovation roundup for neurosurgery, published every Sunday for the week before; every item links to its sources. Latest neurosurgery news · All specialties