IODA: An input/output deep architecture for image labeling

作者：

Highlights：

• IODA is a neural network architecture that directly links a whole image to a whole label decision map.

• It is based on a pretraining of the outputs, i.e. of the label dependencies.

• We outperform the state-of-the-art approach on a real-world medical imaging problem.

• An open source implementation is provided.

摘要

Highlights•IODA is a neural network architecture that directly links a whole image to a whole label decision map.•It is based on a pretraining of the outputs, i.e. of the label dependencies.•We outperform the state-of-the-art approach on a real-world medical imaging problem.•An open source implementation is provided.

论文关键词：Deep learning architectures,Deep neural network,Image labeling,Machine learning,Medical imaging,Sarcopenia

论文评审过程：Received 19 September 2014, Revised 27 February 2015, Accepted 18 March 2015, Available online 27 March 2015, Version of Record 16 May 2015.

论文官网地址：https://doi.org/10.1016/j.patcog.2015.03.017