Beyond the Patterns 31 - Roger Soberanis Uncertainty-based Graph Convolution Segmentation Refinement

0:04 Introduction Roger Soberanis-Mukul
1:30 Start of Presentation
4:13 Idea of Uncertainty-driven Segmentation Refinement using Graph Convolutions
16:06 Measuring Uncertainty in Segmentation
18:42 Graph Convolutions
21:58 Algorithm Overview
44:10 Datasets
45:23 Results
57:05 Detail Analysis of Diversity
1:11:48 Summary and Conclusion
1:13:29 Discussion
1:19:55 Concluding Remarks
It’s a great pleasure to announce an invited talk from TU Munich by Roger David Soberanis Mukul in Beyond the Patterns.
Abstract: Organ segmentation is an essential pre-processing step in different computer-assisted tasks, and currently, deep convolutional neural networks lead the state-of-the-art. However, the nature of the medical images can lead to errors in the segmentation process, generating false negative and false positive regions in the results. Recent works have shown that the uncertainty of deep convolutional neural networks (CNN) can provide helpful insights about potential errors in the network’s predictions. Inspired by these works and the recent graph convolutional networks, we propose using the CNN’s uncertainty to formulate the refinement process as a semi-supervised graph learning problem. To validate our method, we refine the predictions of a 2D U-Net, trained on the NIH pancreas dataset and the spleen dataset of the medical segmentation decathlon. Finally, we perform a sensitivity analysis on the parameters of our proposal.
Short Bio: Roger Soberanis is a PhD student at the Chair for Computer Aided Medical Procedures and Augmented Reality, Technical University of Munich. He studied a bachelor’s in computer engineering and a master’s in computer science at the Mathematics faculty of the Autonomous University of Yucatan, Mexico. His work focus on deep convolutional and graph-convolutional networks for medical applications, with a particular interest in medical image segmentation.
References
Paper www.melba-jour...
This video is released under CC BY 4.0. Please feel free to share and reuse. See "Beyond the Patterns" without ads here: www.fau.tv/cou...
For reminders to watch the new video follow on Twitter / maier_ak or LinkedIn / andreas-maier-a6870b1a6 . Also, join our network for information about talks, videos, and job offers in our Facebook and LinkedIn Groups lme.tf.fau.de/....
Music Reference:
Damiano Baldoni - Thinking of You (Intro)
freemusicarchi...
Damiano Baldoni - Poenia (Outro)
freemusicarchi...