A Multifunctional MRI model Based on IVIM and DKI Predicts HIF-1α, Ki-67, and VEGF Status in Breast Carcinoma

Zhengtong Wang¹Fan Zhao¹Laimin Zhu¹Weiwei Wang¹Wenwen Zhao¹Hao Yu¹Yunxi Li¹Chongchong Li¹Xiuzheng Yue²Yueqin Chen¹Zhanguo Sun^1*

• ¹Affiliated Hospital of Jining Medical University, Jining, China
• ²Philips Healthcare, Beijing, China

Purpose: The research aims to explore the predictive significance of diffusion-weighted imaging (DWI), diffusion kurtosis imaging (DKI), intravoxel incoherent motion (IVIM), and their integrated models in relation to Hypoxia-inducible factor-1 alpha (HIF-1α), Ki-67, and vascular endothelial growth factor (VEGF) expression levels in breast carcinoma. Materials and Methods: This retrospective study included 104 patients with pathologically confirmed breast carcinoma from our institution as the training set, while an external validation cohort of 91 eligible patients was recruited from another tertiary medical center. Two independently working radiologists analyzed IVIM-derived parameters apparent diffusion coefficient (ADC), true diffusion coefficient (D), perfusion-related diffusion coefficient (D*), and perfusion fraction (f), and DKI-derived parameters mean diffusivity (MD) and mean kurtosis (MK). ROC curves were constructed for evaluation of diagnostic efficacy. The outcomes of the multivariate logistic regression model were employed to create a nomogram of the combined model for molecular marker status prediction. Results: High expression levels of HIF-1α, VEGF, and Ki-67 were consistently associated with lower D, MD, and ADC values, and higher perfusion-related D*, f, and MK values (all P＜0.05). ROC curve analysis showed that among the individual parameters, the D value exhibited the highest predictive efficacy (AUC=0.724). ROC curve analysis revealed that the f parameter was the most powerful single indicator for predicting VEGF expression (AUC=0.882). ROC curve analysis indicated MD as the most predictive single parameter for Ki-67 expression (AUC=0.762), showing significantly greater effcacy than D* (Z=2.022, P=0.043). In the training set, the combined models integrating select parameters from IVIM and DKI showed significantly higher predictive performance (AUCs: 0.852-0.923) compared to individual parameters. This performance was replicated in the external validation set (AUCs: 0.841-0.918), with no statistically significant difference in AUCs between the training and external validation sets according to DeLong's test (all P > 0.05). Moreover, the solid line provided a better approximation of the ideal dotted line, indicating higher predictive accuracy of the nomograms (P=0.59, 0.40, and 0.08). Conclusion: Our findings suggest that IVIM may be usefully combined with DKI to help predict the expression levels of Ki-67, HIF-1α, and VEGF in breast cancer, generating hypotheses for future research.