The immune microenvironment of tumors provides information on prognosis and prediction. A prior validation of the immunoscore for breast cancer (ISBC) was made on the basis of a systematic assessment of immune landscapes extrapolated from a large number of neoplastic transcripts. Our goal was to develop a non-invasive radiomics-based ISBC predictive factor.
Immunocell fractions of 22 different categories were evaluated using CIBERSORT on the basis of a large, open breast cancer cohort derived from comprehensive information on gene expression. The ISBC was constructed using the LASSO Cox regression model derived from the Immunocell type scores, with 479 quantified features in the intratumoral and peritumoral regions as observed from DCE-MRI. A radiomics signature [radiomics ImmunoScore (RIS)] was developed for the prediction of ISBC using a random forest machine-learning algorithm, and we further evaluated its relationship with prognosis.
An ISBC consisting of seven different immune cells was established through the use of a LASSO model. Multivariate analyses showed that the ISBC was an independent risk factor in prognosis (HR=2.42, with a 95% CI of 1.49–3.93; P<0.01). A radiomic signature of 21 features of the ISBC was then exploited and validated (the areas under the curve [AUC] were 0.899 and 0.815). We uncovered statistical associations between the RIS signature with recurrence-free and overall survival rates (both P<0.05).
The RIS is a valuable instrument with which to assess the immunoscore, and offers important implications for the prognosis of breast cancer.