PV26 Speakers

Subject to change.

 

 

image

Maria Amezquita, MD

PGY-2 Pathology Resident, Wake Forest

Maria Alejandra Amezquita, MD, is a PGY-2 Pathology resident at Wake Forest Baptist Medical Center, trained at Icesi University in Colombia. With six peer-reviewed publications in high-impact journals, her research spans lung cancer biomarkers, rare disease diagnosis, and AI-driven imaging analysis. Dr. Amezquita is dedicated to advancing diagnostic medicine through innovative, evidence-based approaches that improve patient outcomes.

 

 

SESSIONS

Governance readiness of pathology AI publications: a systematic literature analysis
   Sat, Oct 17
   3:35 PM - 3:55 PM PT
  Seaport H

Introduction: The Joint Commission and Coalition for Health AI released Responsible Use of AI in Healthcare (RUAIH) guidance in September 2025, defining seven elements for responsible clinical AI deployment. The extent to which pathology AI publications document these governance elements has not been measured.

Methods: We searched PubMed (2019-2026) using ('computational pathology' OR 'digital pathology') AND (artificial intelligence OR deep learning OR machine learning), yielding 1,460 publications across 368 journals. We operationalized the seven RUAIH elements into the Governance Readiness Score (GRS): a 0-2 anchored rubric per element (maximum 14) built on explicit textual indicators and rubric-aligned keyword ontologies. A scripted GPT-4o pipeline with a fixed rubric prompt scored title/abstract records for all 1,460 publications (interim analysis); full-text re-scoring of the PubMed Central subset (n=942) is underway. A blinded 30-paper subset was independently rescored by a second large language model (Pearson r=0.96; 90% within 1 point). A board-certified pathologist blindly scored a stratified 10-paper subset (5 highest, 5 zero) with substantial expert-model agreement (Cohen's κ=0.79 expert-vs-primary; κ=0.80 expert-vs-secondary; Landis-Koch).

Results: Mean GRS was 0.42/14 (3%). 940 of 1,460 publications (64.4%) scored zero on all seven elements; the maximum observed was 8/14. Only the risk and bias assessment showed substantive engagement (28.8% of publications scored>=1); safety event reporting was nearly universally absent (99.9% zero). No abstract referenced an AI model card.

Conclusions: Pathology AI abstracts rarely document governance scaffolding expected for responsible clinical deployment under RUAIH. The most actionable gaps were safety event reporting and model-card-level transparency. Governance-aware publication standards are needed before clinical deployment scales.

Learning Objectives: 

  1. Describe seven RUAIH elements relevant to safe clinical deployment of pathology AI tools.
  2. Apply the Governance Readiness Score to evaluate pathology AI publications for deployment readiness.
  3. Identify governance documentation gaps relevant to authors, reviewers, and health systems.
Chat bot