SegAN: Adversarial Network with Multi-scale L1 Loss for Medical Image Segmentation
Brown Bag Lecture by Dr. Sharon Xiaolei Huang | 9/19/2017 11:30AM – 12PM | 7th Floor Conference Room, Bldg 38A
In the field of medical image segmentation, deep convolutional neural networks (CNN) have been applied with promising results. Existing such methods utilize a pixel-wise loss such as softmax in the last layer of their networks, which is insufficient to learn both local and global contextual relations between pixels. In this talk, we introduce an end-to-end Adversarial Network architecture, called SegAN, for segmentation. SegAN consists of a fully convolutional neural network as the segmentor and an adversarial network with a novel multi-scale L1 loss function as the critic. The segmentor and critic networks are trained in an alternating fashion in a min-max game. We show that the proposed SegAN framework is more effective and stable for the segmentation task and it leads to better performance than several other deep CNN based segmentation methods.
Sharon Xiaolei Huang is an Associate Professor in the Computer Science and Engineering Department at Lehigh University. Her research interests are in the areas of biomedical image analysis, computer vision, and computer graphics. She develops novel and robust algorithms for image-based screening and diagnostic tests, image segmentation, registration, and pattern recognition. In these areas she has a track record of publications in highly selective conference proceedings and journals. She has received a P.C. Rossin Professorship at Lehigh University and a Lindback Minority Junior Faculty Award for Career Enhancement. Her research has been funded by the NIH, NSF, the Pennsylvania State, and Air Products. She received her Ph.D. degree in Computer Science from Rutgers University at New Brunswick.