147P Racial bias in pretreatment MRI radiomics features to predict response to neoadjuvant systemic treatment in breast cancer: A multicenter study in China, Germany, and the US

• DOI: 10.1016/j.esmoop.2024.103134
• Publication Date: 2024-05-17T11:19:33Z
• Authors: A. Pfob; J. Liu; C. Sidey-Gibbons; J. Nees; F. Riedel; B. Schaefgen; R. Togawa; J. Huang; J. Heil; B. Gao; M. Golatta; L. Cai
• Keywords: Systemic therapy; Neoadjuvant Therapy

Machine learning with radiomics showed great potential to predict response neoadjuvant systemic treatment (NAST) for breast cancer. However, performance across different ethnicities and racial bias remain unclear. We aimed develop an intelligent algorithm using pretreatment MRI in addition clinical variables, validate their ethnically diverse populations. used institutional data of patients who underwent before NAST. developed a support vector machine based on German features patient tumor variables pCR status (ypT0 ypN0). N4 field correction maintain the consistency images acquired by types machines settings. Model was validated American Chinese dataset. Findings were compared histopathologic evaluation surgical specimen. The main outcome measure area under curve (AUC). included 656 development set, 88.6% (581 656) white people 34.8% (228 achieved pCR. model good set (AUC: 0.81, 95%CI, 0.71-0.83). Validation population sample (n = 100, 80% people) non-inferior 0.75 vs. p 0.543), but 0 decreased 0.61 0.004). Also within training descriptively lower Asian (n= 59, AUC 0.73 (95% CI 0.56-0.88)), or African (n=16, 0.71 0.44-0.95)) ethnicity. Racial exists studies should be assessed global application AI-based imaging algorithms. Ethnic diversity is crucial mitigate when developing such