|
Early Lung Cancer Detection from Pulmonary CT Scans
Automated pulmonary nodule detection involves identifying pulmonary nodule locations from whole lung CT scans using computer methods without any intervention of a radiologist or other personnel. With the recent wide availability of high resolution imaging modalities like helical CT scanners, there is a great need of automated detection systems. With higher resolution scans, the radiologist must visually inspect a large number of slices making it a very tedious and time consuming task. The detection procedure output can be integrated with characterization and diagnosis procedures building up a complete computer aided diagnosis system. There are many issues addressed in our detection algorithm design. Pulmonary nodules have a large variation in size and attenuation characteristics. Hence, our detection algorithm has been designed to accommodate this variation. Consideration of the effect of digitization in detection has been integrated in the design. This is due to the dependence of the image representation of the nodule on the acquisition modality parameters. The other phenomenon of great concern is occurrence of nodules attached to other pulmonary structures. These pulmonary structures could be blood vessels or pleural surface. The detection algorithm has also been designed to detect nodules in these circumstances. Our technique is centered on nodule detector design based on the classification of nodule types. Individual models were designed for each nodule type considering their shape and intensity characteristics. The detection algorithm uses a 3D multiple region filter bank. The filter bank follows a cascaded pattern with sequentially size increasing filters. This concept of cascaded filter is useful to detect wide size range nodules in CT scans. |