中文摘要：

为了实现CT计划图像中前列腺的自动分割,提出一种基于群体CT计划图像的多任务前列腺分割方法.将群体CT计划图像分别映射到不同参考图像空间,形成多个训练任务.利用随机森林算法和自动上下文模型训练出一系列随机森林分类器,将分类器作用在待分割CT计划图像上获得多个分类概率图,最后使用多数投票法求得最终分割结果.实验表明,与单任务分割方法相比,基于群体CT图像的多任务分割能有效提高CT计划图像中前列腺的分割准确率.

英文摘要：

To automatically and accurately segment prostates in CT planning images,a multi-task CT prostate segmentation method is proposed based on group images.The group images with those from other patients are first mapped to various spaces of reference images to form a multiple training task.The random forest method and the automatic context model are used to train a series of classifiers.The trained classifiers are then iteratively applied to CT images to be segmented.Multiple classification probability maps are thus produced.The final segmentation result is obtained using a majority voting method.Experimental results show that,compared with single-task segmentation,proposed multi-task segmentation based on group images can effectively improve accuracy of prostate segmentation for CT planning images.