A cerebral aneurysm is an abnormal dilatation of an intracranial artery that affects about 3–5% of adults. Most aneurysms occur near the arterial bifurcations of the Circle of Willis, particularly in the anterior communicating artery (Acom), posterior communicating artery (Pcom), middle cerebral artery (MCA), and internal carotid artery (ICA) bifurcations.They are usually asymptomatic, but when they rupture, they can lead to a subarachnoid hemorrhage (SAH), a sudden, severe, and often fatal brain hemorrhage with up to 40-50% mortality.Ruptured cerebral aneurysms remain the most common cause of non-traumatic SAH, making early detection and monitoring essential to prevent devastating outcomes.

Imaging Modalities and Their Limitations

The main imaging tools used to detect cerebral aneurysms are:

• CT Angiography (CTA): fast, widely available, and high-resolution; however, it involves radiation and iodinated contrast, and small aneurysms near bone can be missed.
• MR Angiography (MRA): non-invasive and avoids radiation, but has lower spatial resolution and can produce flow or motion artifacts. Sensitivity for aneurysms <5 mm can be as low as 35%.
• Digital Subtraction Angiography (DSA): the gold standard, offering the best resolution and simultaneous treatment potential, but it’s invasive and time-consuming.

Even with these modalities, detection remains challenging. Around 10% of aneurysms are still missed in routine reads. Overlaps on MIP (maximum intensity projection) images, fatigue, and satisfaction of search all contribute to diagnostic misses.

Because of these limitations, AI-based Computer-Assisted Detection (CAD) tools have been developed to support radiologists, improving sensitivity and workflow efficiency.

Study 1: Di Noto et al., 2025- Assessing workflow impact and clinical utility of AI-assisted brain aneurysm detection: A multi-reader study (NeuroImage: Clinical) https://www.sciencedirect.com/science/article/pii/S2213158225001056

Purpose:

To determine whether AI-assisted detection improves radiologist accuracy and workflow efficiency using Time-of-Flight MR Angiography (TOF-MRA).

Methods:

• 460 TOF-MRA scans (360 training, 100 testing)
• 3D U-Net AI model trained on multicenter ADAM dataset
• 2 radiologists (junior + senior) reviewed 100 test scans with and without AI assistance
• Metrics: sensitivity, specificity, reading time

Results:

• AI model: Sensitivity 74%, Specificity 95%, 1.6 false positives per case
• Junior reader: Sensitivity ↑ 78 → 83%, Specificity ↓ slightly
• Senior reader: No major change
• Reading time: Slightly increased by ~15 seconds per case

Clinical Takeaway:

AI assistance did not significantly improve diagnostic accuracy and slightly reduced specificity for less experienced readers. However, it highlights how workflow optimization is needed before full clinical integration.

Study 2: Din et al., 2023 — Detection of cerebral aneurysms using artificial intelligence: a systematic review and meta-analysis (J NeuroIntervent Surg)

https://pubmed.ncbi.nlm.nih.gov/36375834/

Purpose:

To evaluate the accuracy of AI algorithms for aneurysm detection on CTA, MRA, and DSA, compared with or without radiologist assistance.

Methods:

• 43 studies (2004–2021), 18,143 patients
• Followed PRISMA-DTA and QUADAS-2 guidelines
• Included 34 standalone AI and 9 AI + radiologist studies

Results:

•  Pooled sensitivity and specificity showed strong overall performance, but variability across studies and patient populations remained high.

Clinical Takeaway:

AI demonstrates strong potential in aneurysm detection, but false positives and study bias remain concerns. It should support, not replace, radiologists until validated in large, multicenter real-world studies.

Study 3: Hu et al., 2024 A deep-learning model for intracranial aneurysm detection on CT angiography images in China: a stepwise, multicentre, early-stage clinical validation study (Lancet Digital Health)

https://www.thelancet.com/journals/landig/article/PIIS2589-7500(23)00268-6/fulltext

Purpose:

To evaluate whether a deep-learning model could detect aneurysms on CTA with accuracy comparable to radiologists and improve workflow efficiency.

Methods:

• Multicentre prospective study across 27 hospitals in Chinaams
• Compared AI alone, radiologists alone, and AI-assisted radiologists

Results

Clinical Takeaway:

The deep-learning model significantly improved detection accuracy and efficiency, especially when used as an assistant. It was well accepted (92.6% adoption) and maintained high performance even in real-world hospital workflows.

Study 4: Zhou et al., 2024 Classification, detection, and segmentation performance of image-based AI in intracranial aneurysm: a systematic review -(BMC Medical Imaging)

https://pubmed.ncbi.nlm.nih.gov/38956538/

Purpose:

To summarize the performance of AI models in aneurysm classification, detection, and segmentation across imaging modalities.

Methods:

• 47 studies (up to March 2024)
• Models categorized as:
  • Classification (21)
  • Detection (4)
  • Segmentation (12)
  • Hybrid (10)
• Metrics included accuracy, sensitivity, specificity, Dice, and mIoU

Results:

• Classification models: Accuracy 74–99%, Sensitivity up to 99%
• Detection models: Sensitivity 82–94%, Specificity up to 96%
• Segmentation models: Dice 22–95%, mIoU up to 96%
• Hybrid models: Accuracy 79–94%, Sensitivity up to 97%

Clinical Takeaway:

Deep learning and hybrid AI models achieve high diagnostic accuracy, but inconsistent datasets and limited clinical validation still hinder widespread adoption. Both Dice and mIoU scores measure how well the AI’s segmentation overlaps with a radiologist’s; higher scores mean better accuracy.

Conclusion

Across all four studies, one message is clear: AI has tremendous potential to enhance aneurysm detection, support radiologists, and improve workflow efficiency. Deep-learning models, especially those tested in large multicentre environments, show that AI can safely act as a second reader, improving both speed and accuracy. But before widespread clinical use, we still need more standardized datasets, explainable models, and prospective validation. As AI continues to evolve, it’s becoming a valuable partner, not a replacement, in the early detection and prevention of cerebral aneurysm rupture