A wavelet frames + K-means based automatic method for lung area segmentation in multiple slices of CT scan

Computer assisted detection of lung nodules offers a more accurate method of nodule detection which leads to reliable diagnosis of lung cancer. Lung segmentation is a first step in the process of automatic detection of nodules. In this paper, we propose a wavelet packet frames based approach for effective lung segmentation. The proposed algorithm selects the optimal wavelet representation that is a collection of wavelet packet frames. The frames are subsequently used for clustering of coefficients using k-means clustering, which leads to the segmented lung region. The algorithm is tested on the one publicly available dataset containing 350 images and CT scan dataset of 5 local patients containing a total of 71 images. Accurate segmentation of lung is acquired with average difference in pixels from the ground truth being as low as 1.34±0.451. Furthermore, the proposed technique is fully automated and is capable of segmenting lung in multiple slices with no manual intervention or change in parameters.