(Sunday, 7/14/2019) 5:00 PM - 6:00 PM
Room: 221CD
Purpose: Cancer treatment strategies have been decided according to lung cancer stages, and thus different radiomic prognostic estimation models should be investigated for the stages. We aimed to explore the feasibility of different prognostic prediction models for lung cancer stages I-IIIB based on radiomic signatures.
Methods: 204 positron emission tomography-computed tomography (PET-CT) images of lung cancer patients (adenocarcinoma, squamous cell carcinoma, or large cell carcinoma) with cancer stages I (n = 38; 18.6%), II (n = 22; 10.8%), IIIA (n = 52; 25.5%), and IIIB (n = 92; 45.1%) were chosen from The Cancer Imaging Archive (TCIA). 216 radiomic features were computed from histogram and texture matrices after applying two-dimensional wavelet decomposition filters to the images. A different radiomic signature for each stage was constructed from seven features selected by using elastic-net-regularized cox proportional hazard models (CPHMs). P-values in likelihood ratio test for CPHMs with radiomic signatures based on all patients and separately constructed for stages I, II, IIIA and IIIB patients were compared to evaluate the impacts of cancer stages. The smaller p-value indicates a better model in prediction of patients’ prognoses.
Results: The p-values of CPHMs for stages I, II, IIIA, IIIB and all patients were 1.4×10��, 5.0×10�³, 7.8×10��, 0.15 and 0.19, respectively. Stage-based CPHMs showed smaller p-values than all-patients-based CPHMs.
Conclusion: The prognostic prediction model for each lung cancer stage could be feasible in decision-making of treatment strategies prior to the cancer treatment.
Keywords
Image Analysis, Wavelets, Texture Analysis
Taxonomy
IM- CT: Quantitative imaging/analysis
Contact Email
