Automatic honeycombing detection using texture and structure analysis

Honeycombing in the lung is an important diagnostic sign for diseases involving fibrosis of the lung. Furthermore, the quantification of honeycombing is needed to determine the severity of the disease. In this paper, we present a novel method of automatically detecting honeycombing regions in high resolution computed tomography images of the lung. We detect potential honeycombing cysts within the lung boundary and cluster them based on Euclidean distance. The texture attributes of the cluster region are then calculated. We also use the regional information of the cluster as honeycombing occurs predominantly in the peripheral region of the lung. This regional information has not been used in any of the literature reported and allows us to distinguish honeycomb cysts from other similar looking structures such as the bronchi. A decision tree is generated using the Weka J48 algorithm, with the training examples supplied by the radiologist. The decision tree is then used in the automatic classification of honeycombing regions. The classification performance is evaluated by comparing against the honeycombing regions provided by the radiologist.