医学图像分割--Segmentation of Medical Ultrasound Images Using Convolutional Neural Networks

Segmentation of Medical Ultrasound Images Using Convolutional Neural Networks with Noisy Activating Functions

http://cs229.stanford.edu/proj2016/report/Li-SegmentationOfMedicalUltrasoundImagesUsingConvolutionalNeuralNetworksWithNoisyActivatingFunctions-report.pdf

这是 Stanford University 一个学生做的 project

使用 U-Net 做超声图像分割，主要改进的地方是使用 Noisy Activating Functions 激活函数
关于Noisy Activating Functions激活函数参考文献：
Noisy Activation Functions
https://arxiv.org/abs/1603.00391v3

超声图像还是比较难分割的：

人也很容易分割错误

这里采用 U-Net 网络结构：

直接将 U-Net 网络 用于超声图像分割，效果不是很好
In order to improve the performance of it, we explore the possibility to use the noisy activation functions to push the algorithms out of local minima and improve its segmentation accuracy.

The noise serve to push the algorithm out of local minima and make the algorithm explore a larger area
加入噪声希望算法可以跳出局部极小值