Triple-Negative Breast Cancer and Artificial Intelligence: Current Paradigms in Diagnosis, Therapy Prediction, Prognosis, and Challenges

Triple-negative breast cancer (TNBC) is an aggressive subtype of breast cancer that has a distinct histopathological profile, heterogeneity, and poor prognosis, with a limited number of available therapies. Artificial intelligence (AI), machine learning (ML), deep learning (DL), and radiomics have fundamentally improved the accuracy of diagnosis, prognosis, and therapy in TNBC. Recent AI advances in TNBC include transformer-graph convolution models like NACNet (90% accuracy, 96% sensitivity, AUC 0.82) for NAC response prediction, and longitudinal deep learning radiomics models achieving AUCs of 0.924 (training) and 0.875 (testing) by integrating ultrasound and clinical data. Hybrid CNN-Bi-LSTM-EfficientNet mammography models have reached 99.2% accuracy, while ConvNeXtBase ultrasound models achieved 89% accuracy and F1-scores of 0.81 for TNBC subtype classification. AI is also being paired with nanotechnology to create intelligent drug-delivery systems, reduce toxicity, predict drug resistance, and integrate tumor microenvironment, immune biomarkers, and radiomics for personalized therapy. AI-based imaging models have shown excellent accuracy in terms of diagnostics and subtyping of TNBC, with AUC values reaching up to 0.97. In the same context, DL-based models based on whole-slide histopathology images and radiomics predicted response to neoadjuvant therapy with AUC values as high as 0.96. AI-derived immune infiltrating cell (IIC) signatures, radiomics-derived omics models, and spatial tumor microenvironment (sTME) traits have been demonstrated to be prognostics for disease-free survival (DFS) and overall survival (OS) as well. Various AI-based prognostic models, including the AI-based TLS-TB index nomogram, clinic-radiomics signatures, and Digi-sTILs scores, have C-index values from 0.65 to 0.76, supporting moderate to strong prognostic classification. Incorporating AI-derived immune signatures, tumor microenvironment, and radiomics with clinical parameters facilitates personalized planning for risk stratification, prognostic prediction, and selection of treatment options for management of TNBC compared to the traditional TNM staging alone. However, the generalizability of AI-derived models continued to be tested due to variations in training datasets and the biological heterogeneity of TNBC, which challenged their implementation and clinical applicability. Therefore, multi-center validation and prospective clinical trials will be the need of the hour to fully integrate AI as a tool in the standard practice of precision oncology and personalized management of TNBC.