Machine Learning-Based Integration of DCE-MRI Radiomics for STAT3 Expression Prediction and Survival Stratification in Breast Cancer

Dong Pan^1,2,3Cheng-Yan Zhang^4,5,6Ya-Fei Wang^5,6Shuang Liu^1,2,3Xiongzhi Wu^1,2,3,5,6*

• ¹Tianjin Nankai Hospital, Tianjin Medical University, Tianjin, China
• ²Tianjin Key Laboratory of Acute Abdomen Disease Associated Organ Injury and ITCWM Repair, Tianjin, China
• ³Institute of Integrative Medicine for Acute Abdominal Diseases, Tianjin, China
• ⁴Department of Gastroenterology,Shanxi Hospital Affiliated to Cancer Hospital, Chinese Academy of Medical Sciences; Cancer Hospital Affiliated to Shanxi Medical University, Taiyuan, China
• ⁵Tianjin Medical University Cancer Institute and Hospital, National Clinical Research Center for Cancer, Tianjin, China
• ⁶Tianjin's Clinical Research Center for Cancer, Tianjin, China

Objective: To explore the association between signal transducer and activator of transcription 3 (STAT3) expression, tumor immune microenvironment, and overall survival (OS) in breast cancer, and to develop a non-invasive radiomics model for early risk stratification using dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI).Methods: Data from 1,008 patients with breast cancer in The Cancer Genome Atlas were analyzed to evaluate the prognostic significance of STAT3 expression using Kaplan-Meier survival analysis and Cox regression models. Functional enrichment and immune cell infiltration analyses were performed to assess TIME tumor immune microenvironment characteristics. Additionally, DCE-MRI data from 101 patients in The Cancer Imaging Archive were used were used to extract radiomic features from early-and delayed-phase images. A STAT3 predictive model was developed using six machine learning algorithms. Model performance was assessed using receiver operating characteristic (ROC) and related diagnostic statistical indicators. Results: Low STAT3 expression was significantly associated with poorer OS (hazard ratio [HR] = 1.927, p < 0.001) and an unfavorable tumor immune microenvironment. GSEA revealed that high STAT3 expression enhanced epithelial apoptosis and TNFα/NFκB signaling while suppressing pro-tumorigenic pathways, which was associated with an immunosuppressive microenvironment, whereas low STAT3 correlated with T-cell exhaustion. DIA confirmed elevated STAT3 in tumor versus normal tissue (p < 0.05). The logistic regression-derived radiomics model for STAT3 expression prediction exhibited consistent discriminative performance, with area under curve (AUC) values of 0.861 (95% CI: 0.749 -0.947) in the development cohort and 0.742 (95% CI: 0.588 -0.884) in the validation cohort. High radiomics-derived scores were positively correlated with elevated STAT3 expression, longer OS (p = 0.034), and immune-related gene signatures indicative of a heightened immune response. Conclusion: Radiomics analysis of DCE-MRI images in this study offered a noninvasive method for predicting STAT3 expression and characterization of the tumor immune microenvironment. This approach can offer valuable insights into breast cancer prognosis and support the development of personalized therapies. Clinical Relevance: The developed radiomics model offers a clinically feasible, noninvasive approach for prognostic assessment and may inform personalized immunotherapy strategies by providing concurrent evaluation of STAT3 expression and tumor immune microenvironment characteristics.