Deep Spatiotemporal Convolutional Network for Brain Reperfusion Prediction using Multiphase Computed Tomographic Angiography and Clinical Data

Brain reperfusion injury is a complex and multifactorial phenomenon that occurs when blood supply is restored after a period of ischemia. Reperfusion can be achieved either by endovascular therapy or thrombolytic therapy through mechanical removal of thrombi or tissue plasminogen activator (tPA) and lead to significant neuronal damage and adverse clinical outcomes. Accurate prediction of brain reperfusion injury is crucial for timely intervention and improved patient management. To efficiently predict tissue perfusion and infarction, we propose to develop a multimodal deep framework using multi-phase computed tomographic angiography and clinical data. We have presented a spatiotemporal convolutional network for image feature extraction, which is then combined with clinical features for brain reperfusion prediction. Experiments are conducted on the IACTA-EST challenge test set, which showed our proposed framework achieved significantly better performance and ranked 1 in the IACTA-EST challenge.