1.5 dimensional circular pattern filter for multiscale lung nodule detection

The work presented is an important step in computerized lung nodule detection. The existing schemes proposed in the literature are generally two-step procedures, i.e., segmentation of suspicious areas followed by a differentiation of nodules from lung and chest structures. The algorithms in these schemes for the suspicious area segmentation are all based on 2-D matched filtering techniques which may not be robust in the situations where lung structures overlap with nodules. The 1.5-D circular pattern filter proposed in this paper uses a grouped local gradient characteristics method which is insensitive to overlapped structures, but sensitive to a pre-defined shape, such as circular area. A set of multiscale filters combined with a multiscale stability analysis provides a very effective method for detection of lung modules of different sizes.