<p>This study proposes a novel Mamba-based model, DeepLamba, designed to improve efficiency and accuracy in medical image segmentation tasks. DeepLamba seeks to address the limitations of existing convolutional and transformer-based models, particularly concerning computational complexity, parameter count, and performance on multi-scale features. DeepLamba is an encoder-decoder model that incorporates feature transform processes using Mamba-blocks in the encoder and a lightweight decoder with a receptive field block (RFB) for multi-scale feature manipulation. The model also includes skip connections with squeeze-and-excitation (SE) blocks, enabling channel-wise attention to improve segmentation performance. The model was evaluated on two medical image datasets, including microscopy and abdominal MRI images, using metrics such as mean and median F1-score, Dice similarity coefficient (DSC), and average normalized surface difference (NSD). Empirical evaluations demonstrated that DeepLamba achieves competitive or superior accuracy compared to state-of-the-art models across both datasets while reducing the parameter count by up to 72% and floating-point operations by up to 39%. Ablation studies indicated the effectiveness of the SE-enhanced skip connections and RFB layers in improving segmentation on images with varied scales. DeepLamba offers a parameter-efficient alternative for medical image segmentation, balancing segmentation accuracy and computational resource demands, making it suitable for real-world applications where both performance and efficiency are critical.</p>