医学图像的交互式分割方法相比于自动分割方法在复杂的临床环境中能得到更精确和鲁棒的结果，在放疗靶区勾画、器官建模、手术规划等任务中具有巨大的临床需求。目前的医学图像交互式分割方法普遍面临需要的人工交互多、耗时较长、智能化程度较低的问题，成为其在临床上大规模应用的主要瓶颈。深度学习技术通过从图像大数据中进行学习，自动提取最有效的特征，有望克服以上交互式分割系统中的瓶颈问题。本项目旨在研究一套基于深度学习技术的交互式分割算法，实现分割精度高、响应速度快、人工交互少的智能化医学图像分割方法，将针对以下基础问题进行深入研究：1）深度学习与人工交互相结合的实现机制，2）适用于交互式分割的神经网络模型，3）深度学习对分割结果进行智能化修正的实现机制。本项目有望实现更高效率、更高精度的交互式医学图像分割，为提高临床上计算机辅助诊疗系统的智能化程度和鲁棒性提供算法支撑。

Compared with automatic methods, interactive methods for medical image segmentation can achieve more accurate and robust results in complex clinical environment, and have a large clinical demand in many tasks such as target region delineation for radiotherapy, organ modeling and surgical planning. Existing interactive medical image segmentation methods are commonly faced with requiring a large amount of human interactions, consuming long time and low level of intelligence, which have limited their large scale application in clinic. Deep learning techniques can automatically extract the most useful features by learning from a large set of images, and are expected to overcome these problems in interactive segmentation systems. This project aims to develop a group of new interactive segmentation algorithms to achieve more intelligent medical image segmentation with highly accurate results, fast response, and fewer human interactions. The following fundamental problems will be studied: 1) the mechanism to realize the combination of deep learning and human interactions, 2) neural network models that are suitable to interactive segmentation, 3) the mechanism to intelligently refine segmentation results with deep learning. This project is expected to realize novel interactive medical image segmentation methods with higher efficiency and accuracy, so that to improve the level of intelligence and robustness of computer-aided diagnosis and therapy systems in clinic.