MRI-based radiomics for non-invasive prediction of histopathological and immunohistochemical features in ductal carcinoma in situ: Implications for risk stratification

Background Differentiating the immunohistochemical and histopathological features of ductal carcinoma in situ (DCIS) lesions is critical to potentially enable future tailored treatment options and de-escalation strategies. Purpose To investigate the potential of magnetic resonance imaging (MRI) radiomics for the non-invasive prediction of estrogen receptor (ER), progesterone receptor (PR), grade, microcalcification, and comedonecrosis in DCIS lesions, which may provide implications for risk stratification. Material and Methods This single-center retrospective study included 75 DCIS patients who had undergone pre-treatment MRI and were diagnosed through surgical pathology at our hospital between January 2010 and December 2023. The 75 segmented lesions were categorized into subgroups based on ER, PR, comedonecrosis, microcalcification status, and grade. Model performance was then assessed. Manual segmentation was performed on the second phase of dynamic contrast-enhanced T1 MRI. The BORUTA method was employed for feature selection. Using the selected features, the classification performance of various models was examined to determine if DCIS could be differentiated across the five subgroups. Results In total, 75 patients with DCIS (74 women; mean age=49.36±11.28 years) were evaluated. Effective discrimination in subgroups was as follows: comedonecrosis, n =5; grade, n =1; ER, n =3; and PR, n =5. Radiomic model performance varied across subgroups, with F1 scores observed in the subgroups: ER, 0.8293–0.9091; PR, 0.7796–1; grade, 0.9051–0.9538; microcalcification, 0.9126–0.9895; and comedonecrosis, 0.8247–1. Conclusion MRI radiomics demonstrates significant potential in non-invasively differentiating key immunohistochemical and histopathological features of DCIS. This approach may provide complementary information for DCIS characterization and risk stratification, although further validation is required before clinical implementation.