Page 11 - AIH-1-4
P. 11
Artificial Intelligence in Health Radiomics in early-stage lung cancer
Oncological outcomes Seven AIP radiomics features were associated with DM (CI: 0.638±0.676) LR model AUC (CI): OS: 0.88 (0.78 – 0.97) RFS: 0.86 (0.76 – 0.96) LR-RFS: 0.85 (0.74 – 0.95) AUC values of 0.72±0.04, 0.83±0.03, and 0.60±0.04, for recurrence, local recurrence, and non-local recurrence, respectively CI values for LR in clinical, radiomics, and combined models were 0.57 (0.39 – 0.75), 0.55 (0.38 – 0.73), and 0.61 (0.43
Algorithm/statistical method used Univariate and multivariate analyses LR Multivariable LR models RSF LR, DT, SVM Multiple linear regression Kaplan–Meier analysis and log-rank test
Important radiomics identification method PCA and factor analysis (FactoMineR package) Pearson’s correlation SVD, LASSO Adaptive LASSO LASSO Regression model LASSO
Number of significant radiomics features 19 166 2 16 4 4 (OS) 2 (LRFS) 3 (PFS) 5 Abbreviations: FB: Free breathing; AIP: Average intensity projection; PCA: Principal component analysis; DM: Distant metastasis; CI: Concordance index; LASSO: Least absolute shrinkage and selection operator; RSF: Random survival forest; OS: Overall survival; LRFS: Local relapse-free survival; PFS: Progression-free survival; LR: Logistic regression; DT:
Table 1. SBRT response prediction studies using CT‑based radiomics
Number of radiomics features and radiomics extraction method 644, MATLAB 2013 toolbox 219, Definiens Developer 43, MATLAB 944, PyRadiomics Ver. 2.2.0 1502, PyRadiomics 107, PyRadiomics v3.0.1 432, MATLAB-based Radiomics tools package vector machine; LC: Local control; MFS: Metastasis-free survival; SVD: Singular value decomposition; LR-RFS: Locoregional recurrence-free survival.
Number of patients 112 59 70 573 119 358 125
Imaging method used FB CT and AIP CT Follow-up thorax CT after SBRT FB planning CT Breath-hold planning CT FB planning CT FB planning CT FB planning CT
Study and year Huynh et al., 2017 9 Li et al., 2017 15 Lafata et al., 2019 14 Kakino et al., 2020 10 Luo et al., 2022 12 Sawayanagi et al., 2022 11 Isoyama-Shirakawa et al., 2023 13
Volume 1 Issue 4 (2024) 5 doi: 10.36922/aih.3541
