'Eigen Nodule': View-based Recognition of Lung Nodule in Chest X-ray CT Images Using Subspace Method

摘要

In this paper, we propose a new recognition method of nodules using each CT value itself in ROI (region of interest) area as a feature value. In the learning phase, first, candidate regions are automatically classified into some groups using principal component analysis. A set of CT values in a ROI including each detected region is regarded as a feature vector, and then subspaces are calculated for each group by applying KL expansion. In the testing phase, a testing feature vector is projected to each subspace. Distances from the testing vector to the projected points are measured. If the distance to the nodule subspace is nearest, it is determined to be a nodule. Otherwise, it is determined to be a blood vessel. By applying our new method, the number of the false positives is 0.9[shadow/slice] without false negatives.