Texture analysis of blood flow maps in CT perfusion studies of NSCLC: correlation with the overall survival

Purpose: This study aims to assess the effectiveness of second-order texture features computed on CT perfusion (CTp) colorimetric maps, rather than on the original CTp sequences, as prognostic biomarkers. Methods and Materials: 15 patients with primary lung cancer underwent axial CTp examination and survival data were included in the study. Mean overall survival was calculated at 8.3 months. Blood flow (BF) values were computed using the maximum slope method; values undergoing high fitting errors were automatically removed. Seven global features based on first-order statistics and six local-based second-order statistics were computed on BF maps and k-means clustering algorithm was used to assess their correlation with overall survival (OS). Results: The inverse difference moment and the correlation are the most performing second-order features. In two examinations, the second-order features correlate with OS, where first-order features fail. These perfusion maps clearly show heterogeneity patterns, whose salient features cannot be extracted with simpler global analyses. Values of the couple of the second-order features allow achieving a linear separation between the groups of 10 patients with OS≤8.3 and 5 patients having OS>8.3. Conclusions: These results show that the local-based features, computed on CTp maps where unreliable BF values have been removed, result to be a valuable prognostic factor for OS. In fact, the capability to consider the spatial relationships of perfusion values, thus, the functional heterogeneity characterizing the lesions, makes them really attractive. Accordingly, this technique applied on the texture of perfusion maps represents an encouraging step forwards to the clinical utilization of CTp