193 - Noninvasive Imaging of the Tumor Microenvironment in Hepatocellular Carcinoma Treated with Sorafenib and Memory-like NK Cells

Purpose: Hepatocellular carcinoma (HCC) presents a significant global health burden due to late-stage diagnosis and limited treatment options. Sorafenib provides modest survival benefits in advanced HCC but is often hindered by resistance and toxicity. Memory-like NK (pNK) cells have emerged as promising immunotherapeutic agents due to their potent cytotoxicity functions. This study investigates the therapeutic potential of combining Sorafenib with memory-like NK cells immunotherapy in HCC, with a focus on different administration routes and noninvasive monitoring strategy using MRI.

Materials and Methods: Two rat HCC models (N1S1 and MCA) were used to assess treatment outcome. Animals received control, Sorafenib, pNK cell, or combination therapy. NK cells were administered locally (in N1S1) or systemically (in MCA). Quantitative MRI was used to monitor tumor progression weekly. Histopathological analysis (H&E, CD56, CD31, TUNEL) was used to illustrate tumor microenvironment (TME) characteristics. Random Forest models were trained using MRI-derived radiomic features to predict histopathological changes.

Results: The N1S1 model exhibited significantly greater response to all therapies, particularly combination therapy with local NK cell delivery, which nearly halted tumor growth. Baseline histological assessments revealed higher NK infiltration, vascularization, and rate of spontaneous apoptosis in N1S1 tumors compared to MCA. The machine learning models trained on MRI-derived radiomic features successfully predicted key histopathological features with high accuracy (R² > 0.90). Models based on T1-weighted and T2-weighted imaging individually demonstrated robust accuracy, with their combination offering consistent improvements in prediction.

Conclusion: The enhanced therapeutic response in the N1S1 model is likely driven by its more immunogenic and vascularized TME. Local pNK cell delivery markedly improved efficacy. These findings underscored the value of integrating quantitative MRI with machine learning to non-invasively monitor TME dynamics. This approach holds a promising potential to enhance individualized HCC therapy while mitigating key constraints of conventional diagnostic approaches. Limitations of this study include small sample size, few histopathological markers, and the conventional T1 and T2 imaging. Future studies integrating more diverse datasets, multiple histopathological features, and higher-quality MRI inputs could potentially strengthen the current findings.