DeepLung论文笔记
论文主要思路
DeepLung contains two parts,nodule detection and classification.
3D Faster R-CNN is designed for nodule detection
gradient boosting machine(GBM) with 3D dual path network (DPN) features is designed for nodule classification
论文框架
3D Faster R-CNN for Nodule Detection
The model employs 3 anchors and multitask
learning loss, including coordinates (x; y; z) and
diameter d regression, and candidate box classification.
The numbers in boxes are feature map sizes in
the format (#slices*#rows*#cols*#maps). The numbers
above the connections are in the format (#filters,slices#rows*#cols).
损失函数loss为
pi该i点是否为节点
ti是相关坐标组，d是直径
对ground truth有：
Gradient Boosting Machine with 3D Dual Path Net Feature for Nodule Classification
用到DPN网络模型
整个网络框架为
Gradient Boosting Machine(GBM）是什么？
gradient boosting machine (GBM) is a superior
method to build an advanced classifier from these
features
For nodule detection,we design a 3D Faster R-CNN with U-net-like encoder-decoder structure to detect suspicious nodules. Then we input the detected nodules into the nodule classification network.
The 3D deep dual path network is designed to extract
features. Further, gradient boosting machine with different features combined together to achieve the state-of-the-art performance