Abstract Background: Functional tumor volume (FTV) measured from MRI serves as an imaging biomarker for redirecting treatment in the I-SPY 2 trial. This study aims to validate FTV-based MRI models used to predict favorable response early in treatment. Methods: The study cohort included patients enrolled and randomized to treatment in I-SPY 2 between May 2010 and June 2022. Treatment outcome was assessed by Residual Cancer Burden (RCB) at surgery. Receptor subtype-specific FTV-based models were developed at T2: 12 weeks) to predict the binary outcome RCB-0/I vs RCB-II/III, with RCB-0/I being the positive outcome. FTV at T0: baseline, as well as changes from baseline at T1: 3 weeks (dFTV0₁) and T2 (dFTV0₂), were included as variables in the model. The study cohort was divided into training and test sets balanced for key variables. All combinations of FTV variables were tested in the training set to select the highest performing model within each HR/HER2 subtype for predicting RCB 0/I based on maximizing area under the ROC curve (AUC). Results: The study cohort consisted of 2117 patients in total, divided into training (n=1474) and test sets (n=643). Table 1 shows sample sizes and estimated AUCs in the training and test sets for individual subtypes. AUCs estimated in the test set were lower than AUCs estimated in the training set for all subtypes except HR-/HER2- (Triple Negative). However, these differences did not reach statistical significance. Table 2 shows FTV reduction thresholds generated by maximizing the Youden Index of the ROC curve, specificity (sp), sensitivity (se), PPV, and NPV for predicting RCB 0/I estimated in the training set and in the test set. Conclusion: Our results show comparable performance in the training and test sets, with the highest AUC in the TNBC subtype and the highest PPV in the HR-/HER2+ subtype. The FTV-based models are currently being used in combination with core biopsy as part of the pre-RCB clinical algorithm for shortening therapy in I-SPY 2. Citation Format: W. Li, N. Onishi, J. Kornak, C. Yau, D. M. Wolf, J. E. Gibbs, T. J. Bareng, N. N. Le, L. J. Wilmes, P. Metanat, M. Gibbons, E. Price, B. N. Joe, S. Venters, K. Santos-Parker, B. LeStage, M. J. Magbanua, L. J. van 't Veer, I-SPY2 Imaging Working Group, I-SPY2 Network, A. M. DeMichele, L. J. Esserman, N. Hylton. Validation of MRI predictive models for treatment optimization in the I-SPY 2 TRIAL abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32 (4 Suppl): Abstract nr PD6-03.
Li et al. (Tue,) studied this question.