BROWSE
Detail
Conference Paper
Multi-target Learning on asymmetric U-Net for PNI boundary detection
Cited - time in 
Share 
Share
- Authors
- Sangwon Lee, Youngjae Park, Jinhee Park, Giljin Jang, Hyemi Kim
- Issue Date
- 2021-11
- Citation
- International Conference on Big Data Computing, Applications and Technologies (BDCAT) 2021, pp.1-5
- Language
- English
- Type
- Conference Paper
- Abstract
- Since perineural invasion (PNI) can have a poor prognosis, it is important to find out PNI as early as possible. However, detecting PNI in nerve cells from small scale optical images takes a lot of human efforts. Therefore, detecting perineural invasion with deep neural networks would be efficient and helpful in reducing human labor. In this work, we proposed an asymmetric network that can produce a segmentation map in level 2 resolution from level 0 scale. We also proposed a learning policy that utilizes nerves and tumors as targets for subtasks to help the network find PNI regions. Our method achieved about 0.42 for the evaluation metric over fifteen unseen whole slide images.
- KSP Keywords
- Deep neural network(DNN), Human labor, Level 2, Optical image, Segmentation map, Small-scale, asymmetric networks, boundary detection, evaluation metrics, multi-target learning, nerve cells