Validation of CT Radiomics Models Predicting Local Tumor Progression in CRLM

Overview

This study externally and internally validated CT radiomics models combined with clinical parameters to predict local tumor progression (LTP) after thermal ablation of colorectal liver metastases (CRLM). The combined clinical-radiomics model showed promising predictive performance, supporting its potential use in clinical decision-making to identify patients at high risk for LTP.

Background

Colorectal liver metastases (CRLM) are commonly treated with surgical resection, but not all patients are eligible. Thermal ablation, including microwave and radiofrequency ablation, is an alternative treatment. However, local tumor progression (LTP) after ablation occurs in 6–46% of cases and is challenging to detect early on contrast-enhanced CT due to similar imaging characteristics between post-ablation effects and recurrent tumor. Radiomics analysis of post-ablation CT images offers a novel approach to predict LTP risk, potentially enabling earlier intervention or tailored follow-up.

Data Highlights

The study included internal and external validation cohorts selected with strict inclusion/exclusion criteria to match the original study population. CT imaging was performed on 19 different scanners with standardized contrast protocols. LTP was defined as new tumor foci within 10 mm of the ablation zone on follow-up imaging within 24 months. Follow-up imaging was performed every 3 months in the first year and biannually thereafter up to 5 years. Radiomics features were extracted from the ablation zone and peri-ablational rim on portal venous phase CT 2–8 weeks post-ablation.

Key Findings

• The combined clinical-radiomics model previously developed achieved a concordance statistic (c-statistic) of 0.78 (95% CI 0.65–0.87) in the original cohort using leave-one-out cross-validation.
• External validation was performed using an independent cohort from Erasmus Medical Centre Rotterdam, and internal validation used a later cohort from the original institution.
• Patient selection criteria were carefully matched to the original study to ensure comparability, with minor adjustments to increase sample size in the internal validation cohort.
• CT acquisition protocols were consistent with the original study, using portal venous phase imaging 2–8 weeks after ablation for radiomics analysis.
• LTP detection was based on rigorous imaging follow-up and confirmed by additional imaging modalities when necessary.

Clinical Implications

The validated clinical-radiomics model can help identify patients at high risk for local tumor progression after thermal ablation of CRLM, allowing for timely complementary treatments and potentially improved outcomes. Additionally, patients at low risk might benefit from a less intensive follow-up schedule, reducing healthcare burden and patient anxiety. Incorporating radiomics into routine post-ablation imaging assessment could enhance personalized management strategies.

Conclusion

This study confirms the robustness and generalizability of a combined clinical-radiomics CT model to predict local tumor progression in CRLM after thermal ablation. Such predictive tools hold promise for optimizing post-ablation surveillance and treatment planning.

References

1. Original Study Authors/Year -- Development of Clinical-Radiomics Models for LTP Prediction
2. CIRSE Standards of Practice -- Thermal Ablation Procedures