Assessment of intratumor heterogeneity in non-small cell lung cancer by unsupervised K-means clustering of radiomics features based on multiphase computed tomography images
Year: 2025
Type: article
Study design: Observational Study
Abstract: Background: Intratumor heterogeneity (ITH) is a key determinant of tumor progression, drug resistance, and poor survival in patients with cancer. This study aimed to examine ITH in non-small cell lung cancer (NSCLC) via unsupervised K-means clustering based on multiphase computed tomography (CT) radiomics features. Methods: This study retrospectively included 452 patients with NSCLC across three cohorts from two centers. Multiphase CT images [including unenhanced CT (UECT) and contrast-enhanced CT arterial phase (CECT-AP) and -venous phase (CECT-VP)], clinicopathological characteristics, and short-term immunotherapy responses of patients were collected. A total of 104 radiomics features were extracted from lung cancer regions segmented on CT images. Unsupervised K-means clustering was then employed to quantify ITH in NSCLC, with results visualized through ITH distribution heatmaps. The degree of ITH was quantified and termed ITH-Radscore. We assessed the difference in ITH-Radscores derived from UECT and CECT images. Furthermore, we analyzed the associations of ITH-Radscores with the clinicopathological characteristics and immunotherapeutic efficacy for NSCLC to validate the accuracy and effectiveness of our proposed method. Results: By viewing ITH distribution heatmaps of unsupervised K-means clustering based on multiphase CT images, we identified two distinct high-order ITH imaging patterns, designated as the tree-ring pattern and diffuse pattern. ITH-Radscores for the tree-ring and diffuse patterns were significantly different, with values of 0.643-0.768 and 0.921-0.967, respectively. ITH-Radscores derived from UECT images were significantly higher than those from CECT images (all P values <0.001). ITH-Radscores derived from CECT-AP and CECT-VP showed limited diagnostic value for high-grade patterns in lung adenocarcinoma (P=0.209 and P=0.501, respectively), while those derived from UECT images demonstrated significant predictive capability (P=0.019). ITH-Radscores derived from multiphase CT images were significantly associated with the clinical stage, histological grade, lymphovascular invasion, angiogenesis expression, and short-term immunotherapy efficacy for NSCLC (all P values <0.05). Conclusions: Multiphase CT-based unsupervised radiomics analysis effectively revealed ITH in NSCLC, especially on UECT images. ITH-Radscore demonstrated potential as a noninvasive imaging biomarker associated with the clinicopathological characteristics and immunotherapy outcomes of NSCLC, offering valuable insights for precision oncology.
Authors: Ronghua Wang, Lin Wang, Huijing Feng, Yang Jing, Xinzheng Wang, Jinzhi Fang, Jiangfeng Du, Linning E
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W4415655991
FWCI:
Cites:
Cited by:
Related to: