Medical Matting: A New Perspective on Medical Segmentation with Uncertainty

In medical image segmentation, it is difficult to mark ambiguous areas accurately with binary masks, especially when dealing small lesions. Therefore, a challenge for radiologists reach consensus by using masks under the condition of multiple annotations. However, these uncertain may contain anatomical structures that are conducive diagnosis. Uncertainty introduced study situations. Nevertheless, uncertainty usually measured variances between predictions in trial way. It not intuitive, and there no exact correspondence image. Inspired matting, we introduce matting as soft segmentation method new perspective deal represent regions into scenes, namely matting. More specifically, because available dataset, first labeled two datasets alpha matte. Secondly, methods applied natural suitable scene, so propose architecture generate matte row. Thirdly, map highlight from results improve performance. Evaluated on datasets, proposed model outperformed state-of-the-art algorithms large margin, proved be more efficient labeling form than mask.