Chest CT images analysis based on U-Net architecture and three-dimensional radiomics feature quantification

Precise prevention and early detection of lung cancer and pneumonia face challenges such as tumor heterogeneity and limited diagnostic accuracy. To solve the above key clinical issues, with high-resolution CT and deep learning technologies, we developed an innovative segmentation network and feature extraction algorithm, achieving submillimeter 3D quantitative analysis of pulmonary lesions (ICC>0.93). The approach overcomes critical technical barriers including poor generalizability across multicenter imaging data and information loss in 2D-to-3D conversion. This advancement significantly improves pulmonary lesion analysis efficiency and provides reliable imaging biomarkers for lung cancer screening and respiratory disease monitoring, demonstrating substantial value for advancing precision medicine.