Learning With Context Feedback Loop for Robust Medical Image Segmentation

Deep learning has successfully been leveraged for medical image segmentation. It employs convolutional neural networks (CNN) to learn distinctive features from a defined pixel-wise objective function. However, this approach can lead less output pixel interdependence producing incomplete and unrealistic segmentation results. In paper, we present fully automatic deep method robust by formulating the problem as recurrent framework using two systems. The first one is forward system of an encoder-decoder CNN that predicts result input image. predicted probabilistic then encoded network (FCN)-based context feedback system. feature space FCN integrated back into system's feed-forward process. Using FCN-based loop allows extract more high-level fix previous mistakes, thereby improving prediction accuracy over time. Experimental results, performed on four different clinical datasets, demonstrate our method's potential application single multi-structure outperforming state art methods. With loop, methods now produce results are both anatomically plausible low contrast images. Therefore, interconnected via be efficient analysis.