Robust Lung Segmentation Method from CT Images Using Wavelet Transform and K-Means Clustering

Lung segmentation from CT images is a critical task for early lung disease detection. However, existing methods are often limited by their sensitivity to noise and their inability to identify small and irregular lung structures. Here, we present a novel lung segmentation method using wavelet transform and k-means clustering. The proposed method first applies Gaussian smoothing to the input image to reduce noise. Then, the wavelet transform is used to decompose the image into a set of coefficients. These coefficients are then used to extract feature, which is used to train a k-means clustering algorithm. The clustering algorithm assigns each pixel in the image to one of two clusters: lung or non-lung. Finally, morphological operations are applied to the clustered image to remove noise and artifacts.

This algorithm gives an accuracy of 98.46%, DSC 46.15 %, and JSI 85.91%. The algorithm executed without applying gaussian filter and gives an accuracy of 98.48%, DSC 46.22 %, and JSI 86.18%. The mean values for these metrics were slightly higher indicating a marginal improvement in performance. The proposed method has several advantages over existing methods. It is more robust to noise, it can identify small and irregular lung structures, and it is relatively fast and efficient. Our algorithm can accurately identify and separate lung tissues from lung CT images, which can assist radiologists in diagnosing lung diseases.