Algorithm Detects Bleeding Risk Marker on Brain Scans with High Sensitivity

An automated algorithm can detect the CT-angiography spot sign in patients with acute intracerebral hemorrhage, according to a pilot study published in Frontiers in Radiology by researchers at the University of Ottawa. The spot sign, a small focus of contrast extravasation visible on CT angiography, predicts hematoma expansion and poor outcome after spontaneous bleeding into the brain, but interpreting it requires specialized expertise. The new findings suggest that software could take over this task, potentially standardizing identification of patients at highest risk of worsening bleeding.

Early hematoma expansion is a major determinant of outcome after intracerebral hemorrhage. When a brain bleed grows in the first hours after onset, patients face substantially worse prognosis, which is why imaging markers that flag expansion risk are valuable for guiding treatment decisions. The spot sign has emerged as one of the most informative of these markers, but its usefulness has been limited by the specialized expertise needed for interpretation.

The Ottawa team evaluated the diagnostic performance of an automated detection algorithm, referred to as CARL, using multiphase CT angiography. They analyzed patients who presented with spontaneous intracerebral hemorrhage within 4.5 hours of symptom onset and underwent multiphase CTA. From a prospective registry, fifty scans were randomly selected for the analysis, giving the study its pilot-scale scope.

In the study design, the automated algorithm independently classified each scan as spot sign positive or spot sign negative. An experienced neuroradiologist served as the reference standard against which the algorithm’s performance was measured. This head-to-head comparison allowed the researchers to calculate standard measures of diagnostic accuracy.

The results showed that the algorithm detected the spot sign with a sensitivity of 95.7 percent, meaning it correctly identified nearly all scans that the expert reader classified as positive. Specificity was 74.1 percent, indicating that roughly one in four spot-sign-negative scans was incorrectly flagged as positive. In practical terms, the tool rarely missed a true spot sign but produced a moderate rate of false positives.

The authors conclude that automated detection of the CTA spot sign on dynamic and multiphase CTA is feasible and demonstrates high sensitivity, while acknowledging the moderate specificity. They argue that such performance supports further independent validation of the CARL algorithm before it could be considered for routine clinical use. As a pilot study of fifty randomly selected scans, the analysis provides an early proof of concept rather than definitive evidence of clinical benefit.

If validated in larger and independent cohorts, automated spot sign detection could serve several purposes. The researchers point out that such tools may standardize spot sign identification across centers, support clinical trial enrollment by applying uniform detection criteria, and have the potential to facilitate rapid selection of patients at high risk of hematoma expansion. Imaging-based selection strategies are particularly relevant for hemostatic therapies, where treatment must be delivered quickly to patients most likely to benefit.

The study’s limitations follow from its design. The sample was small, drawn from a single prospective registry, and diagnostic accuracy was benchmarked against one expert neuroradiologist rather than consensus reading by multiple experts. The reported sensitivity and specificity therefore describe performance on this selected set of scans and cannot yet be generalized to all stroke centers or all CTA scanners and protocols.

Nevertheless, the findings mark a step toward automating a decision-support task that has depended on specialized human expertise. Because the spot sign predicts both hematoma expansion and poor outcome, a reliable automated detector could help clinicians and trialists alike identify high-risk patients earlier in the acute phase. The Ottawa group’s results, published as a research report in the journal’s neuroradiology section, set the stage for the independent validation studies that will determine whether CARL can move from pilot testing into everyday stroke imaging practice.

Subject of Research: Automated detection of the CT-angiography spot sign in acute intracerebral hemorrhage

Article Title: Automated detection of the CT-angiography spot sign in acute intracerebral hemorrhage

Article References: Dowlatshahi, D., Chakraborty, S., Ramsay, T., & Hogan, M. J. (2026). Automated detection of the CT-angiography spot sign in acute intracerebral hemorrhage. Frontiers in Radiology, 6, Article 1898570. https://doi.org/10.3389/fradi.2026.1898570

Image Credits: AI Generated

DOI: 10.3389/fradi.2026.1898570

Keywords: intracerebral hemorrhage, CT angiography, spot sign, automated detection, hematoma expansion, neuroradiology, diagnostic accuracy, stroke imaging, CARL algorithm, pilot study, Automated, detection