Biomedical research faces substantial challenges in multimodal image segmentation. In recent years, U-shaped models have made significant advance in medical image segmentation. The UNet architecture, while promising in biomedical segmentation, employs a fixed receptive field through a feature-fused model and attention gates. This study introduces several modifications to the classical UNet architecture by altering the receptive field with a feature-fused model. Additionally, the incorporation of a transfer learning model aims to enhance computational efficiency. The modified UNet architecture, combined with transfer learning, leverages pretrained models to reduce training time and improve performance. The proposed model was trained using a publicly available dataset and achieved a dice coefficient of 0.920 on the CVC Clinic dataset, demonstrating its effectiveness in medical image segmentation. This approach not only addresses the fixed receptive field limitation but also enhance computational efficiency, making it a valuable contribution to the field of biomedical research.