Abstract P1-05-17: Radiomic models based on dynamic contrast-enhanced magnetic resonance imaging predict the immunophenotype reflecting spatial distribution of CD8+ tumor-infiltrating lymphocytes in breast cancer

Abstract

Abstract Purpose: This study aimed to develop noninvasive magnetic resonance imaging (MRI)-based radiomic models to predict immunophenotypes of breast cancer. Methods: A total of 182 breast cancer patients who underwent upfront surgery were analyzed, divided into training (n = 137) and validation (n = 45) cohorts. Dynamic contrast-enhanced (DCE)-MRI was acquired, and immunophenotype was determined on the surgical tumor sections: immune-inflamed (high degree of CD8+ T cells infiltrated), -excluded (CD8+ T cells accumulated at invasive margin but not efficiently infiltrated in tumor bed), and -desert (CD8+ T cells absent within tumor and at margins). Based on 833 radiomic features extracted after manual delineation, the least absolute shrinkage and selection operator method was used to build radiomic models. Results: Our radiomic models from the whole tumor sections showed moderate performance in predicting the immune-inflamed versus non-inflamed tumors, showing the highest AUC values, of 0.659 and 0.671 for training and validation cohorts, respectively. Combining the models, much improved predictability was observed, with AUC values of 0.973 and 0.985 for training and validation cohorts, respectively. Other radiomic features from tumor periphery data discriminated immune-excluded versus immune-desert status, with AUC values of 0.993 and 0.984 for training and validation cohorts, respectively. The combined models were also applicable to predicting immunophenotype for different molecular subtypes, with AUC values ≥ 0.867. Conclusions: By integrating the immunohistochemistry profiles, we established MRI-derived radiomic models to predict the detailed immunophenotype of breast cancer. This study suggests the feasibility of noninvasive assessment of tumor immune status in real-world clinics. Citation Format: Seung Hyuck Jeon, So-Woon Kim, Mirinae Seo, Yu Jin Lim. Radiomic models based on dynamic contrast-enhanced magnetic resonance imaging predict the immunophenotype reflecting spatial distribution of CD8+ tumor-infiltrating lymphocytes in breast cancer [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P1-05-17.