Multiparametric MRI-Based Deep Learning and Radiomics for Predicting Progression-Free Survival Benefit in Patients
Wendi Kang1, Yingen Luo2, Xuan Zhou3
1Department of Radiology, Postdoctoral Station of Clinical Medicine, The Third Xiangya Hospital, Central South University, Changsha, China (W.K); Department of Radiology, The Third Xiangya Hospital, Central South University, Changsha, China (W.K., X.Z., P.R); Department of Interventional Therapy, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Panjiayuan Nanli 17# Chaoyang District, Beijing, China (W.K., Y.L., S.W., H.L., Z.Y.).
Rationale And Objectives:
The non-invasive biomarkers for predicting progression-free survival (PFS) in patients with hepatocellular carcinoma (HCC) treated with immunotherapy and molecular targeted therapy combined with transarterial chemoembolization (IMT-MTT-TACE) are urgently needed to identify individuals who are likely to benefit from this treatment regimen. This study aims to develop a non-invasive imaging biomarker for predicting PFS in patients with HCC receiving IMT-MTT-TACE, leveraging the integration of deep learning, radiomics, and clinical factors.
Materials And Methods:
This study included 180 patients with HCC who were treated with IMT-MTT-TACE at two medical centers. Radiomic features were extracted from six sequences of multiparametric magnetic resonance imaging. Deep learning features were extracted based on the ResNet50 algorithm. A Cox regression combined model was developed by integrating significant clinical, radiomics, and deep learning features. Model performance was evaluated using the concordance index (C-index) and time-dependent receiver operating characteristic (ROC) analysis area under the curve (AUC).
Results:
The C-reactive protein and alpha-fetoprotein in immunotherapy (CRAFITY) score was identified as an independent predictor of PFS (P < 0.05). In three cohorts, the C-index values for the deep learning model were 0.757, 0.751, and 0.744, respectively. The C-index values for the combined model were 0.803, 0.746, and 0.744, respectively. In the time-dependent ROC curve analysis predicting 1-year PFS, the AUC values for the combined model were 0.934 (95% confidence interval [CI]: 0.881-0.986), 0.842 (95% CI: 0.699-0.984), and 0.862 (95% CI: 0.725-0.998). The deep learning-based combined model demonstrated good predictive performance and exhibited strong robustness and generalizability.
Conclusion:
The integration of CRAFITY score, radiomics, and deep learning features contributed to predicting PFS for patients with HCC undergoing IMT-MTT-TACE. This combined model holds promise for enabling precise pretreatment risk stratification and optimizing monitoring protocols, thereby guiding prognosis assessment and individualized clinical treatment decisions.
