Brainomix and Remedy Pharmaceuticals have reported exploratory findings suggesting that AI-assisted analysis of routine brain scans could help identify patients with severe stroke who may benefit from an investigational treatment. The post-hoc analysis of the Phase 3 CHARM trial examined whether acute infarct core volume, measured using non-contrast computed tomography (NCCT), was associated with outcomes among patients receiving intravenous glyburide, marketed in development as CIRARA.
Presented at the World Stroke Congress in Seoul, the analysis used Brainomix 360 Stroke software and its in-house AI Core Lab to examine imaging and clinical data. The results point to a potential role for automated imaging biomarkers in selecting patients for future stroke trials, although the findings are exploratory and do not establish CIRARA’s efficacy.
The distinction matters: the analysis identifies a possible relationship between imaging-defined stroke severity and treatment response, not proof that the software or the drug improves patient outcomes in routine clinical care.
Why the 125 mL threshold matters
In the analysis, Brainomix used automated NCCT analysis to measure acute infarct volume in 428 patients. The researchers also examined CT perfusion (CTP) and diffusion-weighted MRI (DWI-MRI) scans alongside clinical outcomes.
Among 362 patients in the per-protocol analysis, CIRARA showed a borderline association with improved functional outcomes on the modified Rankin Scale (mRS), a measure of disability after stroke. Within a subgroup of 191 patients whose acute infarct core volume was below 125 mL, treatment was significantly associated with a more favorable shift in mRS scores and a greater likelihood of reaching mRS 0–3 at 90 days. PR Newswire+1
A broader modified intention-to-treat analysis of 428 patients found a statistically significant non-linear interaction between infarct volume and treatment allocation. This suggests the relationship between stroke severity and potential treatment response may be more complex than a single cutoff.
However, the findings come from a post-hoc analysis. Subgroup associations can help researchers develop hypotheses, but they do not establish that the treatment is effective for those patients. The proposed volume threshold and other imaging markers need prospective evaluation before they can reliably guide treatment decisions.
A key caveat: the original Phase 3 trial missed its primary outcome
The new findings need to be read alongside the main CHARM trial results. The randomized, placebo-controlled Phase 3 study enrolled 535 patients and was stopped early. In its published analysis, intravenous glibenclamide did not produce a statistically significant improvement in the overall 90-day modified Rankin Scale outcome among patients aged 18–70: the common odds ratio was 1.17, with a 95% confidence interval of 0.80–1.71 and a p-value of 0.42. The investigators said early termination left the study underpowered for definitive conclusions. ScienceDirect+1
ClinicalTrials.gov records the trial as terminated for operational challenges and other strategic considerations, rather than for efficacy or safety reasons. ClinicalTrials.gov
That context makes patient selection a central research question. The new analysis does not overturn the overall trial result; instead, it suggests there may be patient subgroups worth investigating in future studies.
Why AI-assisted NCCT analysis could matter
Non-contrast CT is widely used in acute stroke assessment, while CT perfusion and MRI can provide additional information but may not be available in every hospital or clinical pathway. Automated measurement of infarct volume on NCCT could, if validated, make it easier for more sites to apply consistent imaging criteria when enrolling patients in trials.
Brainomix 360 Stroke already offers AI-supported analysis across several imaging modalities, including tools for CT-based stroke assessment. The company’s research functionality was used in this analysis to quantify imaging biomarkers and connect them with clinical outcomes. FDA-Cleared Stroke AI Platform+1
For pharmaceutical developers, such tools could support more consistent trial stratification, retrospective analysis and the identification of candidate treatment-response biomarkers. For clinicians, the potential benefit is more reliable information to support decisions—not an automated determination that a patient should receive an investigational drug.
The next step is independent validation and prospective testing of the proposed imaging criteria. Until then, the findings should be treated as hypothesis-generating rather than a change in clinical practice.
Market Landscape
AI-assisted stroke imaging is moving beyond automated detection and triage toward quantitative imaging biomarkers that can support clinical research. These tools analyze scans to standardize measurements, help clinicians interpret complex images and explore links between imaging features and patient outcomes.
Brainomix’s analysis illustrates a potential second use for medical imaging AI: helping drug developers identify which patients might respond to a therapy. But an association found in retrospective or post-hoc data is not equivalent to a validated predictive biomarker. Regulatory-grade evidence will require reproducible methods and prospective clinical validation.
Top Insights
- New treatment-response signal: Patients with infarct cores below 125 mL showed significant associations between CIRARA treatment and better functional outcomes in a subgroup analysis.
- Routine imaging potential: Automated NCCT analysis could help researchers assess patient eligibility at hospitals without routine access to advanced imaging.
- Important clinical caveat: The original CHARM trial did not demonstrate a statistically significant improvement in overall 90-day functional outcomes.
- AI’s research role: Automated imaging biomarkers may help pharmaceutical companies analyze trial populations and refine future study designs.
- Further evidence required: The findings are exploratory and need prospective validation before they can inform routine treatment decisions.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
