Identification of Precise 3D CT Radiomics for Habitat Computation by Machine Learning in Cancer

• DOI: 10.1148/ryai.230118
• Publication Date: 2024-01-31T14:51:26Z
• Authors: Olivia Prior; Carlos Macarro; Víctor Navarro; Camilo Monreal; Marta Ligero; Alonso Garcia-Ruiz; Garazi Serna; Sara Simonetti; Irene Braña; Maria Vieito; Manuel Escobar; Jaume Capdevila; Annette T. Byrne; Rodrigo Dienstmann; Rodrigo Toledo; Paolo Nuciforo; Elena Garralda; Francesco Grussu; Kinga Bernatowicz; Raquel Perez-Lopez
• Keywords: Repeatability

Purpose To identify precise three-dimensional radiomics features in CT images that enable computation of stable and biologically meaningful habitats with machine learning for cancer heterogeneity assessment. Materials Methods This retrospective study included 2436 liver or lung lesions from 605 scans (November 2010–December 2021) 331 patients (mean age, 64.5 years ± 10.1 [SD]; 185 male patients). Three-dimensional were computed original perturbed (simulated retest) different combinations feature kernel radius bin size. The lower 95% confidence limit (LCL) the intraclass correlation coefficient (ICC) was used to measure repeatability reproducibility. Precise identified by combining reproducibility results (LCL ICC ≥ 0.50). Habitats obtained Gaussian mixture models data using compared all features. Dice similarity (DSC) assess habitat stability. Biologic correlates explored a case study, cohort 13 CT, multiparametric MRI, tumor biopsies. Results showed poor ICC: median [IQR], 0.442 [0.312–0.516]) against 0.440 [0.33–0.526]) but excellent size 0.929 [0.853–0.988]). Twenty-six precise, differing lesions. (DSC: 0.601 [0.494–0.712] 0.651 [0.52–0.784] lesions, respectively) more than those 0.532 [0.424–0.637] 0.587 [0.465–0.703] respectively; P < .001). In correlated quantitatively qualitatively observed MRI histology. Conclusion on enabled assessment through computation. Keywords: Diffusion-weighted Imaging, Dynamic Contrast-enhanced Radiomics, Unsupervised Learning, Oncology, Liver, Lung Supplemental material is available this article. © RSNA, 2024 See also commentary Sagreiya issue. An earlier incorrect version appeared online. article corrected April 5, 2024.