Blood supply to the brain is secured by an extensive collateral circulation system, which can be divided into primary routes, i.e., the Circle of Willis, and secondary routes, e.g., collaterals from the external to the internal carotid artery and leptomeningeal collaterals. Collateral flow is the basis for acute stroke treatment, since neurones will only survive long enough to be rescued with reperfusion therapies if there is sufficient collateral flow. Poor collateral flow is associated with worse outcome and faster growth of larger infarcts in acute stroke treatment. Therapeutic promotion of collateral flow theoretically offers the chance for outcome improvement, but randomised trials are lacking. The extent of collateral flow is highly variable between individuals. As a consequence, the speeds of infarct growth are highly variable, resulting in varying individual treatment time windows until the whole salvageable tissue has become infarcted. An ideal patient selection for reperfusion therapies should be based on imaging of the salvageable tissue, the so called penumbra. The penumbra can be approximately visualised by computed tomography (CT) and magnetic resonance imaging (MRI), but both methods are significantly inaccurate in about 25% of the patients. There is a need for improved penumbra imaging by CT and MRI, and first studies applying machine learning techniques have shown promising results.