Suggestive Annotation: A Deep Active Learning Framework for Biomedical Image Segmentation

Abstract

Abstract. Image segmentation is a fundamental problem in biomedical image analysis. Recent advances in deep learning have achieved promising results on many biomedical image segmentation benchmarks. However, due to large variations in biomedical images (different modalities, image settings, objects, noise, etc), to utilize deep learning on a new application, it usually needs a new set of training data. This can incur a great deal of annotation effort and cost, because only biomedical experts can annotate effectively, and often there are too many instances in images (e.g., cells) to annotate. In this paper, we aim to address the following question: With limited effort (e.g., time) for annotation, what instances should be annotated in order to attain the best performance? We present a deep active learning framework that combines fully convolutional network (FCN) and active learning to significantly reduce annotation effort by making judicious suggestions on the most effective annotation areas. We utilize uncertainty and similarity information provided by FCN and formulate a generalized version of the maximum set cover problem to determine the most representative and uncertain areas for annotation. Extensive experiments using the 2015 MICCAI Gland Challenge dataset and a lymph node ultrasound image segmentation dataset show that, using annotation suggestions by our method, state-of-the-art segmentation performance can be achieved by using only 50% of training data.

Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation

“…For clustering, the authors use a hybrid method based on K-means and max-cover algorithms. The results for both 2D and 3D datasets suggest that the 1-shot active learning method performs comparably to an iterative alternative by Yang et al (2017).…”

“…Table 3 compares the active learning methods suggested for medical image segmentation. Yang et al (2017) propose a framework called suggestive annotation where the candidate samples for each round of Algorithm 1: Active learning Input : Initial model M 0 , unlabeled dataset U 0 , size of query batch k, iteration times T , active learning algorithm…”

“…The methods suggested by Kuo et al (2018); Yang et al (2017), despite their differences, both employ an ensemble of FCNs to estimate sample uncertainty, which is slow and computationally expensive as one needs to incrementally train an Publication Type…”

“…Sample selection strategy Annotation unit informativeness diversity annotation cost Gorriz et al (2017) iterative whole 2D image Yang et al (2017) iterative whole 2D image Ozdemir et al (2018) iterative whole 2D image Kuo et al (2018) iterative whole 3D image Sourati et al (2018) iterative 2D image patch Mahapatra et al (2018) 1-shot whole 2D image Sourati et al (2019) iterative 2D image patch Zheng et al (2019) iterative 2D image patch Table 3: Comparison between active learning methods for medical image segmentation. The suggested methods differ in terms of the definition of the annotation unit and the criteria by which these units are selected for the next round of annotation.…”

A Weakly Supervised Method for Instance Segmentation of Biological Cells

“…They include prior knowledge in the form of constraints into the loss function for regularizing the size of segmented objects. The work in [11] proposes a way to keep annotations at a minimum while still capturing the essence of the signal present in the images. The goal is to avoid excessively annotating redundant parts, present due to many repetitions of almost identical cells in the same image.…”