Semi-supervised medical image segmentation has emerged as a promising research direction in the medical domain. Existing semi-supervised methods mostly adopt a single model, which is prone to confirmation bias, and make insufficient use of shape prior knowledge, thereby limiting the segmentation accuracy. To address these issues, we propose a semi-supervised framework based on heterogeneous dual teacher-student mutual learning, which comprises two student models with different networks and two corresponding teacher models. Student models exchange pseudo-labels for cross pseudo supervision, effectively mitigating the negative impact of confirmation bias. Each teacher-student pair adopts the Uncertainty Filtering Strategy (UFS) to select high-confidence pseudo-labels, thereby enhancing the quality of pseudo-supervision signals. Student networks incorporate the Dual Shape Awareness (DSA) module, which integrates signed distance map and differentiable morphological operations to capture fine-grained shape information from both geometric features and structural priors. Experiments on the ACDC and PROMISE12 datasets show that our method outperforms many excellent semi-supervised segmentation methods.