3D pulmonary nodules detection using fast marching segmentation

Abstract

Pulmonary nodule detection is an important step in lung cancer detection because nodules are
the alert signal of lung cancer. The early detection of them can hence increase the patient’s
survival rates. This paper proposes an automated computer aided diagnosis system for
detection of pulmonary nodules based on three dimensional (3D) structures. Lung
parenchyma segmentation using fast marching method was employed. A simple thresholding
technique is used to extract candidate nodules from the segmented lung parenchyma. A 3D
image of nodule candidates is reconstructed by mean of stacked 2D images. To find the
connected voxels of a blob, a 3D connected component labelling is used. Features extracted
from each blob are then fed into the classifier. The random forest algorithm has been invoked
for nodule and non-nodule classification. The proposed detection methodology can give the
accuracy of 92%.

Keywords: lung cancer; pulmonary nodule; fast marching; 3D features; random forest
classifier.