<p>Early screening for intestinal polyps is a critical component in the prevention and treatment of colorectal cancer (CRC). However, existing polyp segmentation algorithms remain inadequate in terms of accuracy and model size to meet clinical requirements. Inspired by the rapid attentional control mechanism within the retinal-superior colliculus(SC)-visual cortex pathway of the biological visual system, we propose a highly lightweight polyp segmentation method, LightRGC-SCNet. In the encoder, we design the lightweight receptive field antagonistic feature extraction module (L-RFAFE). This simulates the antagonistic mechanism and spatial scale diversity of retinal ganglion cells (RGCs) receptive fields, enhancing polyp edge contrast and extracting multi-scale spatial features. In the decoder, inspired by the superior colliculus’s spatial attention-guided mechanism, we construct the joint spatial attention guidance DySample upsampling module (JSAG-DySample). By integrating the DySample with the spatial attention guidance mechanism, we achieve an effective balance between spatial information restoration accuracy and model lightweighting. In the output stage, we employ the semantic and spatial information fusion module (SSIF) to fuse shallow spatial information with deep semantic information. The proposed method was validated on the CVC-ClinicDB, Kvasir-SEG and CVC-ColonDB datasets. Experiments demonstrate that LightRGC-SCNet achieves excellent polyp segmentation performance with a highly lightweight network architecture design (only 0.51 million parameters), meeting the polyp segmentation requirements in clinical scenarios.</p>