A new radiomics model leverages CT imaging features to predict the pathological response of initially unresectable hepatocellular carcinoma (HCC) to neoadjuvant conversion therapy. The study, involving 203 patients across two independent centers, demonstrates that the model, particularly when using the Synthetic Minority Oversampling Technique (SMOTE), outperforms the commonly used modified Response Evaluation Criteria in Solid Tumors (mRECIST) for predicting pathological complete response (pCR). This advancement could refine treatment strategies and surgical timing for patients with advanced HCC.
Improving Prediction of Treatment Response in HCC
Surgical resection offers the best chance for long-term survival in HCC patients, but a significant proportion present with advanced-stage disease, precluding immediate surgery. Neoadjuvant conversion therapy aims to downstage tumors, making them resectable. However, accurately predicting which patients will achieve a pCR, a critical factor for prognosis, remains a challenge. Current methods like mRECIST have shown limited agreement with actual pathological outcomes.
This study introduces a radiomics approach, analyzing high-order features from contrast-enhanced CT (CECT) scans to predict postoperative pathological responses. Radiomics can uncover subtle tumor characteristics not visible through standard imaging assessment, potentially offering a more precise prediction of treatment efficacy.
Development and Validation of the Radiomics Model
The study included patients with initially unresectable HCC who underwent hepatectomy after neoadjuvant conversion therapy, such as transarterial chemoembolization (TACE), hepatic arterial infusion chemotherapy (HAIC), tyrosine kinase inhibitors (TKIs), and immune checkpoint inhibitors (ICIs). Radiomic features were extracted from preoperative CECT scans, and a SMOTE radiomics model was developed to address the imbalance in pCR rates. The model's performance was compared against mRECIST in both training and validation cohorts.
The SMOTE radiomics model demonstrated significantly higher predictive performance than mRECIST in both the training (AUC: 0.832) and validation cohorts (AUC: 0.843). Furthermore, the study found that the pathological response predicted by the SMOTE radiomics model was an independent prognostic factor for recurrence-free survival (RFS).
Clinical Implications and Future Directions
The findings suggest that the SMOTE radiomics model could serve as a valuable tool for preoperative risk stratification and treatment decision support in patients with advanced HCC undergoing neoadjuvant conversion therapy. By more accurately predicting pCR, clinicians can better tailor treatment strategies and optimize the timing of surgery.
"The SMOTE radiomics model could serve as a preoperative risk stratification tool for treatment decision support," the authors stated.
The study acknowledges limitations, including its retrospective design and variations in neoadjuvant regimens. Future research should focus on prospective validation in larger cohorts to further establish the model's clinical utility.
