Add to Chrome
Add to Firefox
Add to EdgeEfforts estimation of doctors annotating medical image
Jan 06, 2019
Accurate annotation of medical image is the crucial step for image AI clinical application. However, annotating medical image will incur a great deal of annotation effort and expense due to its high complexity and needing experienced doctors. To alleviate annotation cost, some active learning methods are proposed. But such methods just cut the number of annotation candidates and do not study how many efforts the doctor will exactly take, which is not enough since even annotating a small amount of medical data will take a lot of time for the doctor. In this paper, we propose a new criterion to evaluate efforts of doctors annotating medical image. First, by coming active learning and U-shape network, we employ a suggestive annotation strategy to choose the most effective annotation candidates. Then we exploit a fine annotation platform to alleviate annotating efforts on each candidate and first utilize a new criterion to quantitatively calculate the efforts taken by doctors. In our work, we take MR brain tissue segmentation as an example to evaluate the proposed method. Extensive experiments on the well-known IBSR18 dataset and MRBrainS18 Challenge dataset show that, using proposed strategy, state-of-the-art segmentation performance can be achieved by using only 60% annotation candidates and annotation efforts can be alleviated by at least 44%, 44%, 47% on CSF, GM, WM separately.
