Lung nodule assessment and risk stratification: performance and variability in low-dose CT screening
Lung nodule assessment and risk stratification: performance and variability in low-dose CT screening
This thesis of Yifei Mao evaluates how lung nodules are assessed and how risk stratification takes place in baseline low-dose CT (LDCT) lung cancer screening. Attention is paid to technical and population-specific factors that are important for clinical practice. It appears that technical factors do not always influence the performance of AI/CAD software in the same way.
Differences in segmentation algorithms between suppliers and variations in slice thickness between 1–2 mm have only a limited influence on volumetric classification at scan level, suggesting that the consequences for CT management decision-making are minor. Within a single AI system, the choice of reconstruction kernel does influence the automatic classification of the type of lung nodule. Therefore, a medium-soft kernel is recommended to ensure consistent and reproducible results. Secondly, we found that Lung-RADS performance is population-dependent. Our meta-analysis shows significant differences in diagnostic accuracy between populations with varying risk profiles (high-risk groups versus general population) and different geographical regions (Asia versus Western countries). This underlines that uniform global application of Lung-RADS is not optimal and that adaptation to the specific population is necessary. Thirdly, CT characteristics of nodules associated with malignancy appear to differ little based on smoking status. This suggests that smoking history alone may not be a reliable predictor of malignancy-related imaging features in nodules, especially in populations with low smoking intensity. Future risk models should therefore also integrate non-smoking-related risk stratification and further refine the management of subsolid nodules.
In conclusion, this thesis emphasizes two priorities for LDCT screening: the systematic validation of AI/CAD systems and population-specific adaptation of management protocols, in order to enhance the effectiveness of screening worldwide.