Lightseg: Efficient Yet Effective Medical Image Segmentation

While recent development in deep learning-based medical image segmentation has been fascinating, effectiveness mostly comes with the expense of expensive computing resources. In search of more affordable and convenient solutions, we propose a lightweight and faster yet effective medical image segmentation approach namely LightSeg. LightSeg leverages separable convolutional layers to decrease the model parameters and an attention mechanism to maintain segmentation quality. Our experimental evaluations on two different backbone networks (U-Net and ResU-Net) in segmenting the lungs from two publicly available chest X-ray datasets demonstrate the robustness of LightSeg while substantially reducing the network parameters.