Boosting Trust in AI Pathology With a New Checklist

As artificial intelligence continues to revolutionize scientific research, its application in pathology is rapidly expanding. However, with this growth come significant challenges regarding the transparency and reproducibility of study findings. To address these concerns, a new set of guidelines has been introduced to standardize how AI-based research is reported.

The Challenge of Reproducibility in AI Pathology

The increasing use of AI tools for automated image analysis (AIA) in pathology has raised questions about the consistency and reliability of published results. Without clear and detailed reporting of the methods used, it becomes difficult for other researchers to validate, replicate, or build upon previous work. This lack of transparency can hinder scientific progress and delay the translation of promising AI tools into clinical practice.

A New 9-Point Reporting Standard

In a recent article published in Veterinary Pathology, an interdisciplinary team of veterinary pathologists, machine learning experts, and journal editors has developed a 9-point checklist to improve the quality of scientific reporting. This checklist provides a clear framework for authors, outlining the essential methodological details that should be included in any manuscript using AI for image analysis. Key areas covered by the guidelines include:

• Dataset creation and composition
• Model training procedures
• Performance evaluation metrics
• User interaction with the AI system

The primary goal is to foster clear communication of methods while actively working to reduce both cognitive and algorithmic bias in research.

Fostering Transparency and Trust

The authors of the guidelines emphasize that transparent reporting is the cornerstone of reproducibility. They argue that making supporting data publicly available is critical for the scientific community. "Transparent reporting is critical for reproducibility and for translating AI tools into routine pathology workflows," the authors state. This includes sharing training datasets, source code, and model weights, which are all essential for enabling meaningful validation and encouraging the broader application of these powerful tools.

These new guidelines are designed to assist authors, reviewers, and editors alike. They will be especially valuable for those submitting to Veterinary Pathology's upcoming special issue focused on artificial intelligence.

Source and Further Reading

The full reporting guidelines were published by Bertram, C. A., et al. in the journal Veterinary Pathology. You can access the full publication for a detailed look at the 9-point checklist. Read the complete journal reference.