Clinical Scene Segmentation with Tiny Datasets

Smith, Thomas J.; Sharkey, Don; Crowe, John; Valstar, Michel

Abstract

Many clinical procedures could benefit from automatic scene segmentation and subsequent action recognition. Using Convolutional Neural Networks to semantically segment meaningful parts of an image or video is still an unsolved problem. This becomes even more apparent when only a small dataset is available. Whilst using RGB as the input is sufficient for a large labelled dataset, achieving high accuracy on a small dataset directly from RGB is difficult. This is because the ratio of free image dimensions to the number of training images is very high, resulting in unavoidable underfitting. We show that the addition of su-perpixels to represent an image in our network improves the semantic segmentation, and that superpixels can be learned to be detected by Convolutional Neural Networks if those superpixels are appropriately represented. Here we present a novel representation for superpixels, multi-channel connected graphs (MCGs). We show how using pre-trained deep learned superpixels used in an end-to-end manner achieve good semantic segmentation results without the need for large quantities of labelled data, by training with only 20 instances for 23 classes.

Citation

Smith, T. J., Sharkey, D., Crowe, J., & Valstar, M. (2019, October). Clinical Scene Segmentation with Tiny Datasets. Presented at 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), Seoul, Korea (South)

• Presentation Conference Type: Edited Proceedings
• Conference Name: 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)
• Start Date: Oct 27, 2019
• End Date: Oct 28, 2019
• Acceptance Date: Aug 20, 2019
• Online Publication Date: Mar 5, 2020
• Publication Date: 2019-10
• Deposit Date: Nov 1, 2019
• Publicly Available Date: Nov 1, 2019
• Pages: 1637-1645
• Series ISSN: 2473-9944
• Book Title: 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)
• ISBN: 978-1-7281-5024-6
• DOI: https://doi.org/10.1109/ICCVW.2019.00203
• Public URL: https://nottingham-repository.worktribe.com/output/3010129
• Publisher URL: https://ieeexplore.ieee.org/document/9022549
• Related Public URLs: http://iccv2019.thecvf.com/; https://ieeexplore.ieee.org/Xplore/home.jsp