Motivation: Radiomics has shown potential in predicting immunohistochemistry in patients with breast cancer, but struggles to fully capture tumor heterogeneity and increase its interpretability. Goal(s): To explore the value of intravoxel incoherent movement-based habitat imaging in predicting immunohistochemistry in patients with breast cancer. Approach: We developed models to predict HER2, Ki-67, hormone receptors (HR), and lymph node metastasis (LNM) of 299 patients with breast cancer including whole-tumor, habitat, conventional MRI feature (CF) and hybrid models. Results: ROC analysis showed that habitat models outperformed whole-tumor models, and combining habitat and CF features yielded the most accurate predictions. Impact: Besides the prediction results, habitat imaging can better depict the heterogeneity of tumors with clearer biological significance of features used, which might help radiologists to accept or reject model's prediction on the immunohistochemistry more readily.
Wang et al. (Tue,) studied this question.