Guided M-Net for High-Resolution Biomedical Image Segmentation with Weak Boundaries

Biomedical image segmentation plays an important role in automatic disease diagnosis. However, some particular biomedical images have blurred object boundaries, and may contain noises due to the limited performance of imaging device. This issue will highly affects segmentation performance, and will become even severer when images have to be resized to lower resolution on a machine with limited memory. To address this, we propose a guide-based model, called G-MNet, which seeks to exploit edge information from guided map to guide the corresponding lower resolution outputs. The guided map is generated from multi-scale input to provide a better guidance. In these ways, the segmentation model will be more robust to noises and blurred object boundaries. Extensive experiments on two biomedical image datasets demonstrate the effectiveness of the proposed method.
© Springer Nature Switzerland AG 2019.