A deep learn-based algorithm for the identification of suspicious local blocks in Lung CT slices : Deep learning method to identify suspicious lung edge

Accurately extracting lung regions from CT slices is an essential task for CT-based lung cancer diagnosis. To deal with this problem, we proposed a deep learning-based neural network to identify the local blocks that contain suspicious missing regions. The proposed method uses a series of image procession methods to extract the preliminary lung region. Then, the scale-invariant feature transform method is utilized to detect key points of the binary image. Forth, the boundary key point method is calculated to replace non-boundary sift key points. After that, the support boundary detecting method is utilized to find the edge of suspicious regions. The minimum enclosing rectangle of the supporting edge is seen as the local block where the suspicious part stays. When all suspicious local blocks are detected, a convolutional neural network is used to classify the suspicious local blocks. Experiment results show that the proposed network can correctly recognize 91.7% of blocks category. The precision value, the loss function value and the receiver operating characteristic curve also prove that the proposed method is an efficient tool.