Abstract
Background. In this study we tried to assess the impact of lung cavities and bronchiectasis on the quality of AIbased emphysema analysis. Methods. A retrospective analysis of chest CT of 50 patients with emphysema combined with lung cavities and bronchiectasis was performed. All studies were evaluated on the same machine with standard technical parameters. Each examination underwent AI-based lung segmentation process and also was assessed by two independent radio logists for visual correctness. Thresholds of –950 HU and –930 HU were used for emphysema evaluation. Results. Programs A and C was capable of defining emphysematous changes starting from 0.2% and program B from 0.3%. Differences in program calculations in one patient ranged from 0 to 17.6%. In 49 out of 50 patients we found bronchiectasis which was included in the final AI-calculations in 100% when analysed by all three programs. Lung cavities were present in 19 out of the 50 patients and in most cases they were considered by programs as areas of emphysema, yet slightly better results were given by program B. A significant overstatement of the estimated emphysema volume presented in program B calculations was discovered, while the results of programs A and C fell within the confidence interval. Conclusions. Lung cavities and bronchiectasis in complex with emphysema significantly affect the result of AI-based analysis. When comparing three software products, there was found a significant overestimation by program B, with a good correlation between programs A and C.
