Automatic method to support radiologists in choosing the most representative slices in CT perfusion of lung cancer

Purpose: At present, several studies still rely on perfusion analysis computed on a single slice. This work provides an effective method to automatically choose the most representative slice, retaining the maximum information content. Methods and Materials: 26 datasets referring to 12 patients with NSCLC, who underwent axial CT perfusion (CTp), were included in this study. Two experienced radiologists chose in agreement the reference slices and manually delineated the lesion, automatically tracked throughout the CTp sequence. Blood flow (BF) values were computed for each slice of lesions and values undergoing high fitting errors were automatically removed. Global entropy (E) is well-known to represent an information measure: it was computed for each BF map, aiming at quantifying functional information, here represented by hemodynamic heterogeneity. All the BF maps of each lesion were ranked for entropy in decreasing order and submitted to the clinician analysis. Results: E ranges from 5.7 to 8.9. The experiments confirm that perfusion slices with the highest heterogeneity are the most representative ones for clinical assessments. Four times they coincide with the reference slice (E=7.0±1.0), nine times are directly adjacent (E=7.6±0.6). Altogether, the most representative slices have entropy 7.2% greater than the reference ones. Conclusion: Several criteria have been adopted to choose the most appropriate slice, mostly based on original CT sequences. This approach may work as an objective, computer aided, system acting as a second reader, this representing a useful tool for more aware clinical considerations. Further analyses including entropy can be used to support visual assessment.