Dr Howard on Current Limitations and Areas of Opportunity for AI in Breast Pathology

"We're seeing kind of this next generation of pathology-based artificial intelligence that I think in several studies clearly can provide more prognostic value than standard genomic tests."

Frederick Howard, MD, an assistant professor of medicine at the University of Chicago, discussed the current landscape of artificial intelligence (AI) in breast pathology, highlighting a field that is rapidly expanding from simple workflow optimization to advanced prognostic modeling.

The current landscape of AI in oncology features various commercially available tools designed to automate and standardize essential pathologic assessments, Howard began. These include applications that quantify immunohistochemistry (IHC) for critical biomarkers such as estrogen receptor, progesterone receptor, and HER2. Additionally, AI is increasingly used to screen biopsy and lymph node resection specimens for the presence of cancer, a capability that significantly reduces the time-intensive burden on pathologists and makes clinical workflows more efficient, he said.

However, Howard emphasized that the true area of opportunity lies in the "next generation" of AI, which seeks to improve upon the current standards of care rather than just increasing efficiency. Recent studies indicate that these sophisticated, pathology-based AI models can provide prognostic value that exceeds that of standard genomic tests, he added. Although currently in the developmental phase, future research aims to refine these tools so they can provide predictive insights into chemotherapy benefit, allowing for highly personalized treatment plans, Howard asserted. A key challenge in this evolution is the transition from generating a "single risk estimate" to providing a comprehensive profile that includes both long-term prognosis and specific markers of response to targeted therapies, he noted.

Advancing this technology further is primarily limited by the challenge of access to increasingly large and diverse datasets, Howard explained. Moreover, Howard pointed out that the field now demands a higher level of scientific rigor. To achieve high-impact clinical integration, models built on available data must be validated through rigorous, blinded assessments to ensure they perform reliably in real-world scenarios, he stated. By overcoming these hurdles of data access and validation, next-generation AI has the potential to fundamentally transform breast pathology from a descriptive science into a more powerful predictive tool, Howard concluded.