Multi-Resolution CNN and Knowledge Transfer for Candidate Classification in Lung Nodule Detection
This paper aim to detect and classify the lung cancer using CNN algorithm.
The hand-crafted features are often used in tradition methods for nodule auto classi?cation. These features may include those primary level features like the texture, shape and size of lung nodules. And it may also include the high level characteristics abstracted from primary level features such as small waves, Local Binary Patterns (LBP) and Histogram of Oriented Gradients (HOG) etc. The features used are self designated, lack of self learning ability like computers and the traditional methods are not intelligent enough to work in an end to end way
In this proposed system the method using convolution neural networks (CNN) is used. As a more automated approach, the CNN method uses raw image data as input and can be directly classi?ed as output. The network can map the image of lung nodule candidate into characteristics of different resolutions and scales while encountering the difficulty in characteristics description of pulmonary nodule for its radiological heterogeneity and variability in sizes and shapes, thus greatly reducing the false positive of classi?cation task and improving the scores on classi?cation metrics.