Optimizing lung volume segmentation by texture classification

Accurate and automated Lung Field (LF) segmentation in volumetric computed tomography protocols is highly challenged by the presence of pathologies affecting lung borders, also affecting the performance of computer-aided diagnosis (CAD) schemes. In this work, a three-dimensional LF segmentation algorithm adapted to interstitial lung disease patterns (ILD) patterns is presented. The algorithm employs k-means clustering followed by a filling operation to obtain an initial LF order estimate. The final LF border is obtained by an iterative support vector machine neighborhood labeling of border pixels based on 3D texture features. The proposed method is evaluated on a dataset of 10 cases spanning a range of ILD patterns such as ground glass, reticular, and honeycombing. The accuracy of the method is assessed using area overlap and shape differentiation metrics (d mean , d rms , and d max ), by comparing automatically derived lung borders to manually traced ones by a radiologist, and further compared to a Gray Level Thresholding-based (GLT-based) method. The proposed method demonstrated the highest segmentation accuracy, overlap=0.942, d mean =1.835 mm, d rms =1.672 mm, and d max =4.255 mm, which is statistically significant (two-tailed student's t test for paired data, p mean =2.324 mm, d rms =3.890 mm,and d max =2.946 mm. The proposed segmentation method could be used as an initial stage of a CAD scheme for ILD patterns.