Medical image segmentation plays a vital role in disease analysis and diagnosis. In recent years, as medical imaging technology has gradually moved from laboratory to clinic, Point-of-Care Testing (POCT) mode has gradually emerged, which can complete detection and analysis in real time around patients and provide timely support for clinical decision-making. However, limited computing resources and hardware conditions in low-resource scenarios such as POCT make it difficult to deploy medical image segmentation networks with large parameters and high computational complexity. Although some lightweight networks are able to run on such platforms, their segmentation accuracy is often difficult to meet clinical requirements. To address this challenge, this paper proposes a lightweight medical segmentation network, LiteMedUNet. LiteMedUNet uses a symmetric encoder-decoder architecture. In the encoding stage, the Perceptual Extraction Attention (PEA) module is integrated to simulate the initial screening process of human eyes and effectively extract the features of potential lesions. In the decoding stage, the Confirmatory Focus Attention (CFA) module is used to enhance and preserve the lesion details while gradually restoring the spatial resolution. The experimental results show that LiteMedUNet achieves high-precision medical image segmentation under the condition of 0.12 M parameters and 2.54 GFLOPs computational complexity, which is an ideal choice for low-resource scenarios such as POCT.