AI, HER2-Low, and the Future of Precision Oncology

In this episode of MD Newsline, Dr Frederick Howard, breast oncologist at the University of Chicago and leader of a research group focused on AI-driven biomarker development, explores the rapidly evolving role of artificial intelligence in breast cancer care. Dr. Howard provides a comprehensive overview of how AI is being integrated into oncology—from radiographic imaging and digital pathology to clinical decision support and language models. He discusses both the promise and the practical challenges of deploying AI tools in real-world workflows, including validation standards, regulatory guardrails, and ethical considerations. The conversation also dives into emerging applications such as AI-based HER2 quantification, recurrence risk prediction from H&E slides, and the potential for multimodal models to transform precision medicine. Episode Highlights: AI in Radiology and Mammography Dr. Howard explains the evolution from early computer-aided detection systems to modern deep learning algorithms trained on millions of mammograms. He discusses emerging AI-driven breast cancer risk prediction tools derived directly from imaging and how they may enhance early detection strategies. Digital Pathology and Biomarker Development AI tools are increasingly capable of quantifying immunohistochemistry and identifying features beyond human visual interpretation. Dr. Howard highlights research presented at major oncology meetings demonstrating improved concordance in HER2-low classification and improved reproducibility in biomarker scoring. Predicting Recurrence Risk Without Genomic Testing One of the most promising areas involves AI models trained on H&E slides to predict recurrence risk—potentially matching or exceeding established genomic assays such as Oncotype DX and MammaPrint. Dr. Howard discusses the validation challenges required before these tools can replace or complement genomic testing in clinical practice. HER2-Low Classification and Antibody-Drug Conjugates The discussion explores limitations of traditional HER2 immunohistochemistry, especially at the lower end of expression. AI-based quantitative approaches may improve patient stratification for HER2-directed antibody-drug conjugates, though questions remain about predictive thresholds and biological mechanisms. Language Models in Oncology Practice Dr. Howard examines the growing use of large language models for literature review, documentation support, and clinical trial matching. He emphasizes the need for HIPAA-compliant systems, clinician oversight, and standardized evaluation frameworks to ensure safe and responsible deployment. Ethics, Governance, and Over-Reliance From data privacy to clinical accountability, the episode addresses the ethical considerations surrounding AI in cancer care. Dr. Howard cautions against over-reliance on AI systems and underscores the importance of maintaining clinician expertise and critical thinking. The Future: Multimodal AI and Precision Medicine Looking ahead, Dr. Howard envisions a future where digital pathology, genomics, imaging, and clinical data converge into multimodal AI systems capable of delivering truly personalized treatment recommendations. He stresses that large-scale data sharing and collaboration will be essential to realizing this potential. Key Takeaway Artificial intelligence is no longer theoretical in oncology—it is actively shaping diagnostics, risk stratification, and treatment selection in breast cancer. However, rigorous validation, ethical governance, and thoughtful integration into clinical workflows are critical to ensuring that AI enhances—rather than replaces—expert clinical judgment. Resources: Website: https://mdnewsline.com/ Newsletter: https://mdnewsline.com/subscribe/ Contact with Dr. Frederick Howard: Here