TNBC NAC Response Prediction via ML and Histology

Pathological complete response (pCR) in triple-negative breast cancer (TNBC) signifies a favorable prognosis, yet only 30-40% achieve it with neoadjuvant chemotherapy (NAC) ) show pCR, leaving 60-70% with residual disease (RD). The tumor microenvironment’s (TME) role in NAC response in TNBC is unclear. Timothy B. Fisher, and his team aimed to introduce a machine learning (ML) two-step pipeline for distinguishing histological components in hematoxylin and eosin (H&E)-stained whole slide images (WSIs) of TNBC tissue biopsies and to identify histological features that can predict NAC response.

Researchers performed an inclusive analysis using H&E-stained WSIs from treatment-naïve biopsies of 85 patients in the model development cohort and 79 patients (41 with pCR and 38 with RD) in the validation cohort, stratified through eightfold and leave-one-out cross-validation. Leveraging a tile-level histology label prediction pipeline and 4 ML classifiers, they scrutinized 468,043 tiles of WSIs. The best-trained classifier employed 55 texture features per tile for probability profiling. Predicted histology classes generated spatial distribution maps. A patient-level NAC response prediction pipeline, trained with features from histology classification maps, employed top graph-based features for radial basis function kernel support vector machine (rbfSVM) classification in predicting NAC treatment response prediction.

The tile-level prediction pipeline demonstrated an accuracy of 86.72% in histology class classification, while the patient-level pipeline achieved a notable 83.53% accuracy in predicting NAC response (pCR vs. RD) within the model development cohort. This model’s validity was substantiated with an independent cohort, showcasing a tile histology validation accuracy of 83.59% and an NAC prediction accuracy of 81.01%. Notably, the histological class pairs with the highest predictive ability for NAC response were tumor and tumor-infiltrating lymphocytes for pCR, microvessel density, and polyploid giant cancer cells for RD.

The study concluded that the ML pipeline robustly identifies clinically relevant histological classes, providing a valuable tool for guiding patient selection in NAC for TNBC.

The study is sponsored by Grants from the National Institutes of Health to RA (R01CA239120). This research was also supported by Georgia State University Molecular Basis of Disease Doctoral Fellowship and the Janssen Scholars of Oncology Diversity Engagement Program awarded to TF.

Source: https://pubmed.ncbi.nlm.nih.gov/38238771/

Fisher TB, Saini G, Rekha TS, et al. (2024). “Digital image analysis and machine learning-assisted prediction of neoadjuvant chemotherapy response in triple-negative breast cancer. Breast Cancer Res.” 2024 Jan 18;26(1):12. doi: 10.1186/s13058-023-01752-y. PMID: 38238771; PMCID: PMC10797728.