UG-Net: Unsupervised-Guided Network for Biomedical Image Segmentation and Classification

Biomedical image segmentation and classification are two critical components in computer-aided diagnosis systems. However, various deep convolutional neural networks are trained by a single task, ignoring the potential contribution of mutually performing multiple tasks. In this paper, we propose an unsupervised-guided network for automated white blood cell (WBC) and skin lesion segmentation and classification called UG-Net. UG-Net consists of an unsupervised-based strategy (US) module, an enhanced segmentation network, and a mask-guided classification network. On the one hand, the proposed US module produces coarse masks that provide a prior localization map for the proposed segmentation network to enhance it in locating and segmenting a target object accurately. On the other hand, the enhanced coarse masks predicted by the proposed segmentation network are then fed into the proposed classification network for accurate classification. Moreover, a novel contextual encoding module is presented to capture high-level information and preserve spatial information. Meanwhile, a hybrid loss is defined to alleviate the imbalance training problem. Experimental results show that our approach achieves state-of-the-art segmentation performance on two public biomedical image datasets.