A recent live stream by Dr. Aleks Digital Pathology & AI, titled "High Concordance, But What Happens in Borderline Cases?," delved into the limitations of artificial intelligence in pathology when faced with ambiguous or borderline diagnostic scenarios. The session, part of the DigiPath Digest #44 series, examined four new PubMed abstracts that explore this critical question.
Key topics included an AI-CAD system for routine gastric biopsy diagnosis, which improved sensitivity for small and dispersed malignant foci, and PD-L1 scoring in non-small cell lung cancer (NSCLC), where high overall concordance still revealed clinically important borderline discrepancies. The discussion also covered lightweight large language models (LLMs) for extracting real-world data from Polish bone sarcoma records.
The stream highlighted that while AI demonstrates strong performance in clear-cut cases, its reliability in borderline situations remains a significant concern for clinical adoption.