From imaging to clinical outcome: dual-region CT radiomics predicting FOXM1 expression and prognosis in hepatocellular carcinoma

Xianyu Chen, Yongsheng Tang, Dong‐Hao Wu, Ruixi Li, Zhiqun Lin, Chao Zhou, Hezhen Wang, Hang Zhai, Junming Xu, Xianjie Shi, Guangquan ZhangFrontiers in Oncology 2023 · DOI

This study, published in Frontiers in Oncology, explores the intersection of medical imaging and molecular biology to improve how we treat liver cancer. By leveraging "radiomics"—the extraction of quantitative data from standard clinical images—the researchers developed a non-invasive method to predict the expression of a key oncogene and patient survival.

TL;DR

Researchers developed a dual-region CT radiomics model that non-invasively predicts the expression of the FOXM1 gene, a major driver of poor prognosis in hepatocellular carcinoma (HCC). By combining these imaging "fingerprints" with clinical data, they created a high-performance nomogram capable of forecasting patient survival at 12, 36, and 60 months.

What Problem Does This Paper Address?

Hepatocellular carcinoma (HCC) is a leading cause of cancer-related deaths worldwide. While traditional markers like tumor stage and alpha-fetoprotein (AFP) levels help, they often fail to capture the molecular complexity of an individual’s cancer.

The gene FOXM1 (Forkhead Box M1) is a known "bad actor" in HCC; its overexpression is linked to cell proliferation, DNA repair, and immune system evasion. However, determining a tumor's genetic profile usually requires an invasive biopsy, which can be risky and may not represent the entire tumor. This paper addresses the need for a non-invasive "virtual biopsy"—using standard CT scans to decode the underlying molecular landscape and predict how a patient will fare.

Methods

The study followed a multi-step pipeline:

  1. Bioinformatics: The team analyzed mRNA data from the TCGA database to confirm FOXM1’s role in HCC, its impact on immune cell infiltration (using CIBERSORTx), and its association with biological pathways (GO/KEGG).
  2. Radiomics Extraction: They analyzed arterial-phase CT scans, extracting 107 features from two distinct areas: the Whole Tumor (WT) and the Peritumoral (WP) region (a 3mm buffer zone around the tumor).
  3. Feature Selection: Using the mRMR (maximum relevance minimum redundancy) algorithm and the Akaike Information Criterion (AIC), they distilled the 107 features down to the most predictive ones (7 for the WT model, 3 for the WP model).
  4. Modeling: They built a Radiomics Score (RS) using logistic regression and integrated it with clinical variables (age, stage, inflammation) to create a prognostic nomogram.

Datasets

  • TCGA-LIHC: 286 patients from The Cancer Genome Atlas were used for the genomic and clinical correlation analysis.
  • TCIA: 35 patients from The Cancer Imaging Archive (with matching TCGA data) provided the CT imaging used to train and validate the radiomics models.

Key Insights and Ideas

  • The WT Advantage: Interestingly, the Whole-Tumor (WT) model significantly outperformed the Peritumoral (WP) model in predicting FOXM1 expression. While the "border zone" of a tumor is often critical in cancer research, the internal architecture of the tumor core proved more indicative of this specific gene's activity.
  • Immune Microenvironment: High FOXM1 expression was linked to an "immunosuppressive" environment—specifically an increase in monocyte infiltration and a decrease in exhausted T-cells, suggesting that FOXM1 helps the tumor hide from the immune system.
  • Oxidative Stress Link: The Radiomics Score (RS) showed a strong positive correlation with oxidative stress genes (like CCNA2 and CDK1). This suggests that the "texture" we see on a CT scan is a macroscopic reflection of microscopic metabolic stress within the cancer cells.

Key Results

  • Predictive Power: The WT radiomics model predicted FOXM1 expression with an AUC of 0.918 in the training cohort and 0.837 in the validation cohort.
  • Survival Stratification: Patients with a high Radiomics Score had a significantly shorter median survival (34.16 months) compared to those with a low score (65.61 months).
  • The Nomogram: The final integrated model (RS + clinical data) was highly accurate at predicting survival probabilities:
  • 12-month survival: AUC 0.892
    • 36-month survival: AUC 0.929
    • 60-month survival: AUC 0.940

Limitations and Future Work

The primary limitation is the small imaging sample size (n=35). While the results are statistically significant, radiomics models typically require much larger, multi-center datasets to ensure they work across different CT scanners and hospital settings.

Additionally, the study is retrospective. The authors suggest that future work should involve prospective clinical trials and deeper experimental validation to uncover the exact biological mechanisms that link FOXM1 activity to the specific visual patterns (features) identified on the CT scans. If validated, this could allow clinicians to tailor aggressive therapies or immunotherapies to patients based on a simple scan.

Summary updated

Abstract

This study highlights the potential of integrating FOXM1 expression and radiomics in understanding HCC's complexity. Our approach offers a new perspective in utilizing radiomics for non-invasive tumor characterization and suggests its potential in providing insights into molecular profiles. Further research is needed to validate these findings and explore their clinical implications in HCC management.
BibTeX
@inproceedings{chen2023from,
  author = {Xianyu Chen and Yongsheng Tang and Dong‐Hao Wu and Ruixi Li and Zhiqun Lin and Chao Zhou and Hezhen Wang and Hang Zhai and Junming Xu and Xianjie Shi and Guangquan Zhang},
  title = {{From imaging to clinical outcome: dual-region CT radiomics predicting FOXM1 expression and prognosis in hepatocellular carcinoma}},
  booktitle = {Frontiers in Oncology},
  year = {2023},
  doi = {10.3389/fonc.2023.1278467},
}
Figure