AI Shows Promise Matching Pathologists at Predicting Triple-Negative Breast Cancer Outcomes

Abstract Background: Tumor-infiltrating lymphocytes (TIL) provide key prognostic information in triple negative breast cancer (TNBC). The CATALINA challenge evaluated multiple TIL-scoring AI algorithms (“cTIL”) on whole-slide images (WSIs) from prospective clinical trial cohorts to assess analytical validity and prognostic performance of cTIL, compared to pathologist-scored stromal TIL (sTIL). Aim: To independently assess the prognostic performance of computational TIL (cTIL) models, compared to pathologist-scored sTIL in a large, prospective cohort. Methods: Two independently developed AI algorithms, producing a total of 5 cTIL scores, were applied to digitized slides blinded to sTIL score and outcomes. We compared agreement between cTIL and sTIL (Spearman’s rho) on 220 breast cancer H&E WSI. We then collected H&E WSI and clinical outcome data for 1,356 early-stage TNBC patients enrolled across seven prospective trials, with pathologist sTIL scored using international guidelines. Five-year invasive disease-free (IDFS), distant disease-free (DDFS), and overall survival (OS) were assessed by Cox proportional hazards model including sTIL and cTIL, adjusted for age, tumor size, nodal status, grade, and treatment. Cases were categorized as high or low sTIL and cTIL, using a previously validated cut-off of 30% for sTIL and the 75th percentile to dichotomize cTIL measures. Discordant cases (cTIL-high/sTIL-low or vice versa) were analyzed for outcome patterns. Results: Moderate correlation (rho 0.37 - 0.47) was observed between cTIL and sTIL scores from the 220 WSI. Pathologist sTIL and all 5 cTIL scores were statistically significantly associated with improved 5-year DDFS, IDFS and OS in the multivariable model. After adjustment for sTIL, only one AI cTIL score (percentage_lymphocyte) measure retained prognostic significance, whereas sTIL remained highly statistically significant for all endpoints (Table 1). Five-year survival probabilities most closely matched the sTIL category. For example, the estimated 5-year DDFS (95% CI) for cTIL/sTIL concordant high/high was 0.82 (0.76 - 0.88), concordant low/low 0.66 (0.63 - 0.70), cTIL-high/sTIL-low 0.67 (0.60 - 0.75), and cTIL-low/sTIL-high groups was 0.78 (0.72 - 0.84), for the percentage_lymphocyte score. Conclusion: In this large, multicenter validation, AI-based TIL quantification demonstrated moderate agreement with pathologist sTIL, and favourable association with prognosis. However, sTIL and cTIL were not directly interchangeable, and discordant cases most closely matched sTIL prognosis. These findings underscore the importance of comparing computational tools to existing, validated biomarkers in large prospective cohorts with comprehensive clinical survival data. Background. Pathologic complete response (pCR) is the absence of residual invasive cancer in the breast and axillary lymph nodes after neoadjuvant therapy. In breast cancer treatment, pCR is a proven surrogate for long-term outcomes. However, accurately predicting pCR at diagnosis remains a clinical challenge. Current tools primarily rely on clinical, genomic, or transcriptomic data. Advances in computational pathology and deep learning enable the extraction of meaningful features from H&E-stained whole slide images (WSIs) to identify phenotypic biomarkers of response. Methods. We apply attention-based multiple instance learning (MIL) to predict pCR from pre-treatment H&E-stained frozen tumor biopsy WSIs in the I-SPY2 trial. A total of 3,306 WSIs from 911 patients across 13 treatment arms were tiled and filtered. Vectors of 1024 features were extracted per tile using the UNI pathology foundation model. The MIL model was independently trained and evaluated with 3-fold cross-validation on these vectors for each arm and tumor subtype. We compared model performance by AUROC between MIL and two elastic net regression models: one trained on pathologist-assessed features (tumor grade, DCIS, invasive histology, and lymphovascular invasion), and another adding clinical features: pre-treatment MRI functional tumor volume (FTV) and transcriptome-derived response predictive subtypes (RPS). Results. 298 of 911 patients achieved pCR (142 HR+/HER2+, 347 HR+/HER2-, 85 HR-/HER2+, and 337 HR-/HER2-). MIL performance varied by arm (AUROC 0.501-0.893) with 6 arms achieving statistically significant performance (95% CI > 0.5). Highest model performance was in HER2+ cohorts: (1) Paclitaxel + Trastuzumab and (2) Paclitaxel + Pertuzumab + Trastuzumab (AUROC = 0.893, 0.785) (Table 1). Of the 6 arms, MIL outperformed the elastic net trained on pathologist-assessed histology features in 5 arms. After including FTV and RPS in the elastic net, MIL still outperformed in 3 arms. Across subtypes, the model predicted better in HR+ subgroups (HR+/HER2- AUROC = 0.706, HR+/HER2+ AUROC = 0.677) than in HR- subgroups (HR-/HER2+ AUROC = 0.533, HR-/HER2- AUROC = 0.548). Conclusion. These findings demonstrate the feasibility of applying MIL vision models to predict treatment-specific response in breast cancer, even with frozen section WSIs and limited data. MIL detects important histology patterns not captured by conventional pathology. Even with added MRI and transcriptomic data, the model provides complementary predictive value. This approach enables early, accurate predictions from routine histology and supports personalized, less toxic treatment—particularly in under-resourced settings. Citation Format: J. R. Dixon-Douglas, D. Drubay, R. Salgado, B. Acs, J. A. van de Laark, Y. Yuan, M. Amgad, L. A. Cooper, Y. B. Hagos, K. AbdulJabbar, J. Meakin, B. Van Ginneken, H. Yan, J. Lemonnier, F. Penault-Llorca, M. Lacroix-Triki, H. Jounsuu, P. Kellokumpu-Lehtinen, S. Loibl, C. Denkert, G. Viale, M. Colleoni, C. Sotiriou, M. Piccart, M. Dieci, S. Demaria, R. Kammler, A. C. Wolff, S. Adams, S. Badve, R. J. Gray, G. Curigliano, A. Vincent-Salomon, T. Nielsen, L. Pusztai, F. Ciompi, S. Michiels, S. Loi. Artificial Intelligence for Tumor-Infiltrating Lymphocytes in Early-Stage TNBC: Results of a Collaborative Prospective TIL Validation Challenge [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 PD11-02.