Artificial intelligence and machine learning applications in the imaging of cerebral vascular lesions: A narrative review

Challenges in cerebral imaging have been present since the dawn of the modern practice of neurosurgery. Despite leaps of progress, subtle pathologies, especially vascular lesions, are still easily missed. In the recent literature, several authors have utilized artificial intelligence and machine learning technşiques to automatically detect and segment various vascular lesions, including aneurysms, AVMs, and cavernous malformations. These models have been trained with computerized angiography (CTA), magnetic resonance angiography (MRA), and digital subtraction angiography (DSA). The models reached the highest sensitivity and specificity in aneurysms due to the simple morphological properties and the abundance of large cıohort datasets. In AVMs, hybrid models incorporating multimodal imaging improve nidus segmentation and radiosurgical planning, though irregular morphology and limited datasets restrict generalizability. Lastly, for cavernous malformations, the models showed promising results, although they lacked external validation. Despite limitations of these models, they consistently reduced diagnostic time and supported clinical decision-making in real-life complex clinical scenarios. Following a series of continuous refinements, their integration into real-world neurosurgical practice is inevitable in the future.