脑梗死是导致人类残疾的主要疾病之一，急性期可通过静脉溶栓及介入机械取栓两种疗法恢复缺血区供血。但目前对于如何选择脑梗死患者进行血管再通治疗的最佳方法还没达成共识，且尚无研究可准确预测患者的预后。近来，人工智能发展迅速，其分支迁移学习是一种将先前建立的模型迁移到新模型中辅助其训练的方法。脑梗死动物模型的构建可人为控制变量，且动物数据的多样性可增加深度学习网络模型的普适性，但从未有研究将动物影像数据建立的模型迁移学习到人类数据。本研究提出以兔的大血管缺血性脑卒中影像数据为基础，开发一种适用于将动物数据建立的模型迁移学习到人类数据的深度神经网络框架，输入原始CT灌注图像对脑组织供血程度进行准确分割，并预测不同临床决策的预后。然后用独立患者队列验证模型的准确性，进而开发出运用于临床的软件。这将有助于临床医生在短时间内科学选择对脑梗死患者预后最有益的临床决策，具有重要临床价值。

Stroke is one of the leading causes for long-term disability in humans. Intravenous r-tPA and mechanical thrombectomy are two methods for restoring blood flow in ischemic regions of the brain. There is currently no clear consensus on the best method for selecting stroke patients to undergo endovascular therapy. Similarly, no prior research has been able to accurately predict prognosis in these patients. Machine learning, in the form of deep learning (artificial intelligence) techniques using deep neural networks, has led to breakthroughs in other areas of visual data processing, and is being increasingly used in medicine. Transfer learning, a machine learning method where a pre-trained model on a problem is reused on a different problem, allows the training of deep neural networks with comparatively little data. Establishment of an animal model of ischemic stroke allows control of variables and increases diversity of training data that improves the generalizability of the model. However, there has been no previous study which transfers learning from model built from animal imaging to human data. In this project, based on imaging data from a rabbit model of large vessel ischemic stroke, we propose to develop a novel deep neural network architectural design that facilitates transfer learning from animal to human data. Using raw CT perfusion images, the model will accurately segment brain perfusion at each voxel (normal vs. penumbra vs. infarct) and predict prognosis after different treatments. We will validate our models on a separate independent patient cohort. A software with an easy-to-use interface will be applied in the ED to generate a report of treatment response prediction and recommend the optimal course of care. This project has great clinical significance because it allows physicians make the most scientific and beneficial decision for stroke patients within a short amount of time.