Treatment Planning in Liver Cancers A surgical focused, up-to-date guide

Why surgeons should care (quick points)

• Non-invasive biology: radiomics converts CT/MRI/PET images into quantitative descriptors that correlate with tumor heterogeneity, microvascular invasion (MVI), and molecular phenotypes and information that traditionally required biopsy.
• Preoperative risk stratification: models using radiomics with clinical data can predict early recurrence, post-hepatectomy liver failure risk, and pathological grade & aiding decisions about extent of resection or need for neoadjuvant/adjunct therapy.
• Treatment selection & targeting: radiomics helps predict which tumors will respond to radioembolization, stereotactic body radiotherapy (SBRT), or ablation and can assist in defining biologically relevant target volumes for image-guided therapies.

How radiomics works (brief, surgical focus)

1. Image acquisition & preprocessing — multiphase CT or MRI (and sometimes PET) with standardized protocols.
2. Segmentation — manually or automatically delineate tumor and peritumoral zones (important for margin decisions).
3. Feature extraction — hundreds of features (shape, intensity, texture, wavelets).
4. Modeling — features are combined with clinical/biomarker data in ML models to predict outcomes (MVI, recurrence, response).
   Surgeons should note: segmentation method and imaging protocol materially affect model performance and reproducibility matters for surgical planning.

Clinical applications relevant to surgical planning

1. Predicting microvascular invasion (MVI)

MVI strongly influences recurrence and guides decisions such as anatomic vs non-anatomic resection and width of margins. Several CT/MRI radiomics models show promising AUCs for preoperative MVI prediction when combined with AFP and clinical features and potentially allowing surgeons to plan wider margins or consider alternative therapies. External validation remains limited.

2. Predicting early recurrence and overall prognosis

MRI-based radiomics has been used to predict early post-resection recurrence and disease-free survival. Predictive tools can influence the choice between resection and liver-directed therapies, and the intensity of surveillance or adjuvant trials. Recent multi-center cohorts and prospective retrospective validations have strengthened evidence but larger prospective trials are still needed.

3. Guiding choice between resection, ablation, and transarterial therapies

Radiomics signatures can differentiate tumors likely to respond poorly to ablation or TACE versus those with favorable imaging phenotypes for local therapies and valuable when balancing operative risk versus minimally invasive options.

4. Target definition for radiotherapy and MR-guided interventions

Radiomics and ML can improve tumor vs background liver discrimination and automate gross tumor volume (GTV) delineation for SBRT and image-guided ablations, improving reproducibility and potentially reducing normal-tissue exposure. MR-guided radiotherapy combined with radiomics enables adaptive strategies in the liver.

5. Predicting post-hepatectomy liver failure (PHLF) and future liver remnant (FLR) outcome

Emerging models combine radiomics with clinical indices to forecast PHLF risk and helpful when planning major hepatectomy or staged resections (ALPPS) and patient selection for preoperative portal vein embolization. These are early-stage but promising.

Recent technical and clinical advances (2023–2025 highlights)

• MRI-based radiomics growth: MRI-radiomics for HCC has expanded with studies showing improved detection, MVI prediction, and recurrence forecasting using multiphase MRI features.
• Integration with MR-guided RT: MR-guided radiotherapy platforms combined with radiomics allow adaptive treatment planning based on imaging biomarkers. Early work shows improved targeting and potential to use radiomics for response prediction.
• Multimodal imaging & PET/CT fusion: combining PET metrics with CT/MRI radiomics increases predictive power in some cohorts (especially for metastatic disease).
• Interpretable ML & survival-path models: newer methods (fusion survival path, explainable models) allow longitudinal prognostication and clearer feature importance & aiding clinical acceptance.

Limitations & practical cautions for surgical teams

• Heterogeneity & reproducibility: differences in scanners, protocols, reconstruction algorithms, and segmentation produce variability. Standardization and harmonization are essential before using models to change surgical plans.
• External validation lacking: many promising models report single-center performance; few have robust multi-center prospective validation required for altering operative strategy.
• Regulatory & workflow integration: regulatory clearance, integration into PACS/OR workflows, and clear responsibility for model outputs are unresolved practical issues.
• Explainability & medicolegal implications: black-box predictions without clear feature explanations are harder to adopt for surgical decision-making.

How to adopt radiomics responsibly in surgical practice (practical roadmap)

1. Multidisciplinary pilot projects: work with radiology, radiation oncology, data science, and IT to pilot validated models within a controlled pathway.
2. Standardize imaging protocols: adopt consistent CT/MR protocols for liver oncology; document reconstruction parameters.
3. Segmentation QA: use validated automatic or semi-automatic segmentation and institute interobserver checks for GTV and peritumoral zones.
4. Use models as adjuncts, not replacements: combine radiomics output with clinical judgment and pathology; reserve major changes in operative strategy for models with strong external validation.
5. Contribute to registries & trials: participate in multicenter validation studies and federated learning efforts to improve generalizability and protect patient privacy.

Near-term research & future directions (what to watch)

• Prospective multicenter trials validating radiomics-guided surgical strategies.
• Federated learning to pool multi-institutional data without moving raw images offsite , accelerates generalizable model training.
• Multi-omics fusion (radiogenomics) linking imaging features with genomics to identify actionable phenotypes and therapy targets.
• MR-guided adaptive therapies where radiomics informs on-table decisions during ablation or MR-LINAC sessions.

Final take (for the operating theatre)

Radiomics is moving from academic promise toward actionable adjuncts that can inform surgical planning for liver cancer , particularly for predicting microvascular invasion, recurrence risk, and treatment response. For now, treat radiomics outputs as decision-augmenting information: useful when validated and interpreted within a multidisciplinary pathway, but not yet a standalone reason to change operative plans without corroborating evidence. Engage with radiology and data scientists to run local pilots, contribute to validation networks, and prepare your unit to safely integrate these tools as the evidence base matures.