Focal FCN: Towards Small Object Segmentation with Limited Training Data
share
Small object segmentation is a common task in medical image analysis. Traditional feature-based methods require human intervention while methods based on deep learning train the neural network automatically. However, it is still error prone when applying deep learning methods for small objects. In this paper, Focal FCN was proposed for small object segmentation with limited training data. Firstly, Fully-weighted FCN was proposed to apply an initialization for Focal FCN by adding weights to the background and foreground loss. Secondly, focal loss was applied to make the training focus on wrongly-classified pixels and hence achieve good performance on small object segmentation. Comparisons between FCN, Weighted FCN, Fully-weighted FCN and Focal FCN were tested on customized stent graft marker segmentation.
READ FULL TEXT
