eduzhai > Applied Sciences > Engineering >

NINEPINS Nuclei Instance Segmentation with Point Annotations

  • Save

... pages left unread,continue reading

Document pages: 7 pages

Abstract: Deep learning-based methods are gaining traction in digital pathology, withan increasing number of publications and challenges that aim at easing the workof systematically and exhaustively analyzing tissue slides. These methods oftenachieve very high accuracies, at the cost of requiring large annotated datasetsto train. This requirement is especially difficult to fulfill in the medicalfield, where expert knowledge is essential. In this paper we focus on nucleisegmentation, which generally requires experienced pathologists to annotate thenuclear areas in gigapixel histological images. We propose an algorithm forinstance segmentation that uses pseudo-label segmentations generatedautomatically from point annotations, as a method to reduce the burden forpathologists. With the generated segmentation masks, the proposed method trainsa modified version of HoVer-Net model to achieve instance segmentation.Experimental results show that the proposed method is robust to inaccuracies inpoint annotations and comparison with Hover-Net trained with fully annotatedinstance masks shows that a degradation in segmentation performance does notalways imply a degradation in higher order tasks such as tissue classification.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×