轮廓指导的层级混合多任务全卷积网络

传统的深度多任务网络通常在不同任务之间共享网络的大部分层(即特征表示层).由于这样做会忽视不同任务各自的特殊性,所以往往会制约其适应数据的能力.提出了一种层级混合的多任务全卷积网络HFFCN,以解决CT图像中的前列腺分割问题.特别地,使用一个多任务框架来解决这个问题.这个框架包括一个分割前列腺的主任务和一个回归前列腺边界的辅助任务.这里,第2个任务主要是用来精确地描述在CT图像中模糊的前列腺边界.因此,HFFCN架构是一个双分支的结构,包含一个编码主干和两个解码分支.不同于传统的多任务网络,提出了一个信息共享模块,用以在两个解码分支之间共享信息.这使得HFFCN可以学习任务的通用层级信息,同时保留一些不同任务各自的特征表示.在一个包含有313个病人的313张计划阶段图片的CT图像数据集上做了详细的实验.实验结果证明了HFFCN网络可以超越现有其他先进的分割方法或者传统的多任务学习模型.

Contour-guided Hierarchically-fused Multi-task Fully Convolutional Network

Conventional multi-task deep networks typically share most of the layers (i.e., layers for feature representations) across all tasks, which may limit their data fitting ability, as specificities of different tasks are inevitably ignored. In this paper, we propose a hierarchically-fused multi-task fully-convolutional network, called HFFCN, to tackle the challenging task of prostate segmentation in CT images. Specifically, we formulate prostate segmentation into a multi-task learning framework, which includes 1) a main task to segment prostate, and 2) a supplementary task to regress prostate boundary. Here, the second task is applied to accurately delineating the boundary of the prostate, which is very unclear in CT images. Accordingly, our proposed HFFCN uses a two-branch structure consisting of a shared encoding path and two complementary decoding paths. In contrast to the conventional multi-task networks, a novel information sharing (IS) module is also proposed to communicate at each level between the two decoding branches, by which our HFFCN endows the ability to 1) learn hierarchically the complementary feature representations for different tasks, and also 2) simultaneously preserve the specificities of learned feature representations for different tasks. We comprehensively evaluated the proposed method on a large CT image dataset, including 313 images acquired from 313 patients. The experimental results demonstrate that our proposed HFFCN outperforms both the state-of-the-art segmentation methods and the conventional multi-task learning methods.