人工智能
计算机科学
正确性
卷积神经网络
功能(生物学)
边界(拓扑)
计算机视觉
分割
图像分割
模式识别(心理学)
对象(语法)
图像(数学)
期限(时间)
像素
算法
数学
生物
物理
量子力学
数学分析
进化生物学
作者
Quanwei Huang,Yuezhi Zhou,Linmi Tao
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2021-04-13
卷期号:: 1798-1802
被引量:6
标识
DOI:10.1109/isbi48211.2021.9433775
摘要
Besides network architecture, researchers have recently focused their attention on the loss function for the Convolutional Neural Network-based medical image segmentation. The loss function is used to evaluate the pixel-wise similarity in the whole image to reflect the correctness of the segmentation, which lacks awareness of the object boundaries. In this work, we propose a novel dual-term loss function with a region-based term and a shape-aware term, which forces the network to learn more boundary details by maximizing the probability difference between the two sides of the boundary. Our method adjusts the output probability appropriately according to the shape. We evaluated our approach on three datasets, and the results show that the proposed loss function outperforms other mainstream loss functions.
科研通智能强力驱动
Strongly Powered by AbleSci AI