Centrala begrepp
The author developed a deep learning framework with domain attention to improve emphysema quantification on CT scans, addressing challenges in large-scale studies and clinical translation.
Sammanfattning
The study focuses on robust quantification of pulmonary emphysema using deep learning techniques. It introduces a novel domain attention block to enhance results by incorporating scanner priors. The research aims to automate and standardize the process for efficient large-scale studies like the MESA Lung Study. By comparing different models, the study demonstrates improved accuracy and generalization with the proposed approach.
Statistik
"average DSC of 66.71% and underestimated the %emph by 0.37% on average."
"achieved an average DSC of 70.23% with an overestimation of %emph by 0.21%."
"regular UNet performed much worse than in-distribution performance with decreased DSC and 127% larger mean error."
"UNet-DAttn (w/ CDFdiff) achieved an average DSC of 60.66% with a mean error of 0.27%."