Semi-supervised learning has made significant strides in the medical domain\nsince it alleviates the heavy burden of collecting abundant pixel-wise\nannotated data for semantic segmentation tasks. Existing semi-supervised\napproaches enhance the ability to extract features from unlabeled data with\nprior knowledge obtained from limited labeled data. However, due to the\nscarcity of labeled data, the features extracted by the models are limited in\nsupervised learning, and the quality of predictions for unlabeled data also\ncannot be guaranteed. Both will impede consistency training. To this end, we\nproposed a novel uncertainty-aware scheme to make models learn regions\npurposefully. Specifically, we employ Monte Carlo Sampling as an estimation\nmethod to attain an uncertainty map, which can serve as a weight for losses to\nforce the models to focus on the valuable region according to the\ncharacteristics of supervised learning and unsupervised learning.\nSimultaneously, in the backward process, we joint unsupervised and supervised\nlosses to accelerate the convergence of the network via enhancing the gradient\nflow between different tasks. Quantitatively, we conduct extensive experiments\non three challenging medical datasets. Experimental results show desirable\nimprovements to state-of-the-art counterparts.\n