| dc.description.abstract | Non-contract low-dose CT of the chest (LDCT) is the standard imaging protocol for lung cancer screening. In addition to lung parenchyma, this imaging modality provides a high-spatial-resolution depiction of the thoracic anatomy, which enables the opportunistic assessment of body composition. In this dissertation, we introduce an artificial intelligence (AI) pipeline for fully automated body composition assessment of lung cancer screening LDCT. The pipeline is developed using the LDCT studies conducted by the Vanderbilt Lung Screening Program. In lung cancer screening LDCT, field-of-view (FOV) tissue truncation beyond the lungs is common. This can affect the accuracy of body composition assessments. To overcome this systematic issue, we propose a semantic FOV extension method that can effectively restore the missing anatomic structures and reduce body composition assessment error introduced by FOV tissue truncation. The proposed two-stage method identifies a new FOV border based on the estimated extent of the complete body and imputes missing tissues in the truncated region. This correction improves both the intra-subject consistency and the correlation with anthropometric approximations. In an analysis based on the National Lung Screening Trial, the body composition measurements automatically derived from the baseline LDCT add predictive value for lung cancer death, cardiovascular disease death, and all-cause death, over 12 years of follow-up. These findings pave the way for increasing the value of population-based lung cancer screening through opportunistic AI body composition assessments. Moreover, our study highlights the potential value of longitudinal assessment of body compositions, particularly in monitoring age-related decline in skeletal muscle health. | |
