Achieving both high accuracy and a lightweight model design is crucial for medical image segmentation, especially for applications requiring efficient terminal deployment. While channel pruning is a well-established approach for model compression, its application to medical image segmentation remains underexplored. This paper introduces a novel model pruning technique called CPG (Channel Pruning with Grouping). Our approach begins by applying a grouping algorithm to pre-arrange the convolutional layers and channels within the network. Subsequently, the \(L_2\) -norm of each channel is computed to assess its significance. Channels deemed less important, along with associated channels within the group, are then pruned to reduce model complexity. Experimental validation on a benchmark dataset confirms the effectiveness of CPG, achieving a 1.30% improvement in Dice score, a 0.80% increase in IoU, and a 93.29% reduction in the number of parameters over the baseline model. These results indicate that CPG significantly decreases model complexity while maintaining accuracy, thereby improving the efficiency and performance of models in medical image segmentation tasks.