The demand for rapid and instant results during medical diagnosis leading to quicker treatment plans (point-of-care) has been drastically increasing. Along this line, we build upon the lightweight UNeXt architecture, another lightweight prototype-augmented segmentation network called UNeXt-Proto that integrates a learnable prototype module controlled by a \(\lambda \_proto\) parameter. We perform an extensive ablation study on \(\lambda \_proto\) to investigate its impact on segmentation performance and computational efficiency. Our evaluation considers both quantitative metrics and model complexity measures, including the number of trainable parameters, GFLOPs, and inference time. Experiments evaluated on the ISIC 2018 and BUSI datasets demonstrate that UNeXt-Proto achieves competitive segmentation performance, with Dice (F1) and IoU scores exceeding UNeXt, while preserving the lightweight property of UNeXt. The \(\lambda \_proto\) ablation reveals the trade-off between prototype strength and feature generalisation, providing insights into optimising prototype integration without compromising efficiency. Benchmarking shows that UNeXt-Proto maintains low computational complexity and fast inference, making it highly suitable for real-world deployment in clinical and resource-constrained environments. Overall, our study highlights the potential of prototype-augmented networks to enhance segmentation quality while retaining the efficiency required for practical point-of-care imaging applications. The code and datasets supporting this work are available from the author upon request via email.