CT Value HistogramBased Classification Framework
Volume measurement is one of several approaches to the management of pulmonary nodules detected by CT scanning [117]. This approach often encounters cases where nodules are volumetrically stable in spite of internal CT value variation. These authors have attempted to develop a 3D computerized method for evaluating the volumetric distribution of CT values within pulmonary nodules. We found that the analysis of CT histograms is a potentially useful method for the quantitative classification of pulmonary nodules without requiring measurement of the proportion of nonsolid and solid components [102, 105, 106]. In [102], we developed a fivecategory classification approach based on the analysis of CT value histograms and investigated the impact of nodule segmentation on classification and the effect of classification on diseasefree survival. We also extended the approach to compute a histogrambased score of recurrence risk to track time interval changes in pulmonary nodules via variational Bayesian mixture modeling for the features obtained from analysis of CT histograms [105, 106]. The key contribution to the computational anatomical models is to represent the internal structure of pulmonary nodules for computing a histogrambased risk score that correlates with prognostic factors. The framework consists of five steps: (1) nodule segmentation, (2) computation of a CT histogram, (3) nodule categorization by applying the variational Bayesian model to cluster CT histograms, (4) computation of the histogrambased risk score by using the combination of the contribution that each category makes to describing the nodule [105, 106], and (5) prognostic prediction using the histogrambased risk score. A schematic overview of the prognostic prediction approach is shown in Fig. 4.5.
