What is it about?
Medical image annotation can be a complex, time-consuming, and error-prone task. This paper examines the problem of imperfect annotations in detecting, segmenting, and classifying multiple instances in histopathology images. It describes a strategy for identifying potentially erroneous annotations in a dataset. The aim is to enhance the reliability of the data for training a deep neural network and improving its efficiency. Our experiments demonstrate the success of this approach.
Featured Image
Photo by Markus Spiske on Unsplash
Why is it important?
Our findings show the value of moving from a large training dataset to a smaller, but more consistently labelled dataset.
Perspectives
This study is a part of the research conducted by the TRAIL Institute (https://trail.ac/en/) in the context of the ARIAC project, which has received support from the Walloon Public Research Service.
Christine Decaestecker
Universite Libre de Bruxelles
Read the Original
This page is a summary of: Training Data Selection to Improve Multi-class Instance Segmentation in Digital Pathology, September 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3632047.3632052.
You can read the full text:
Contributors
The following have contributed to this page
