Discriminative Random Field Segmentation of Lung Nodules in CT Studies
Figure 1
(a) Histogram of voxel intensities of positive examples (nodule voxels) and negative examples (everything else) in training data. Negative examples overwhelm positive examples in all intensities, even in the local area, as shown in this histogram. We must thus exploit locality to achieve a good segmentation. (b) Histogram normalized by number of voxels.