CT Radiomics-based Machine Learning to Identify Intratumoral Fibrosis and Underlying Angiogenesis in Hepatocellular
Tian-Cheng Wang1, Nan Wei2, Yan Bao1
1Department of Radiology, the Second Xiangya Hospital of Central South University, Changsha, No.139 Middle Renmin Rd, 410011, China.
Abstract:
Background Intratumoral fibrosis (ITF) may influence the prognosis of hepatocellular carcinoma (HCC) and transarterial chemoembolization (TACE) efficacy. However, pretreatment biopsies are not routinely performed for HCC. Purpose To develop and validate a CT-based radiomics model to classify ITF grade in HCC and to evaluate the associations of model ITF classifications with post-TACE prognosis and tumor angiogenesis levels. Materials and Methods This multicenter study (August 2013 to November 2023) included five retrospective cohorts and one prospective cohort. Patients in the training (n = 653), internal validation (n = 371), and external validation (n = 173) cohorts were classified as having low or high ITF grade, with histologic Sirius red and hematoxylin and eosin staining as the reference standard. A radiomics model, the Rad-ITF score, was developed and validated to identify ITF grade. In the outcome test cohort (n = 427), patients with intermediate-stage HCC who underwent TACE were grouped by Rad-ITF scores to compare progression-free survival (PFS) and tumor response. Angiogenesis levels of low and high Rad-ITF score in The Cancer Genome Atlas (TCGA) test (n = 39) and genomic test (n = 12) cohorts were determined using RNA sequencing, single-cell RNA sequencing, and spatial transcriptomics analysis. Model performance was evaluated by determining the area under the receiver operating characteristic curve (AUC). Results Overall, 1675 patients (mean age, 54.8 years ± 11.9 [SD]; 1406 male) were included. The Rad-ITF score showed acceptable performance for classifying ITF grade (AUCs: 0.86 [95% CI: 0.83, 0.89], 0.85 [95% CI: 0.80, 0.89], and 0.82 [95% CI: 0.75, 0.89] in the training, internal validation, and external validation cohorts, respectively). Outcome test cohort patients with high Rad-ITF scores (vs low) had a lower tumor response rate (46.2% [66 of 143] vs 62.0% [176 of 284]; P = .002) and worse PFS (6.8 months [95% CI: 5.4, 8.3] vs 11.2 months [95% CI: 10.4, 12.3]; P < .001). High Rad-ITF scores correlated with increased angiogenesis in the TCGA test and genomic test cohorts. Conclusion The Rad-ITF score effectively classified ITF grade and its underlying angiogenesis in HCC and stratified prognosis in patients undergoing TACE. © RSNA, 2026 Supplemental material is available for this article.
