Development of a multi-scanner facility for data acquisition for digital pathology artificial intelligence.

Area of Science:

• Digital pathology
• Artificial intelligence in medicine
• Medical imaging

Background:

• Whole slide imaging (WSI) enables AI in pathology for disease detection and diagnosis.
• AI models are 'brittle,' requiring large datasets and sensitive to input variations from different scanners.
• Current AI training often lacks slide replication across multiple WSI systems, limiting robustness.

Purpose of the Study:

• To establish a multi-scanner facility for comparing scanner performance and replicating digital pathology datasets.
• To improve the robustness and generalizability of AI algorithms in pathology.
• To reduce data requirements for AI training through data emulation.

Main Methods:

• The National Pathology Imaging Cooperative (NPIC) developed the AI FORGE facility at a clinical NHS site.
• The facility integrates 15 WSI scanners from nine manufacturers.
• The process involved planning and constructing a unique multi-scanner environment for data emulation.

Main Results:

• The AI FORGE facility can generate approximately 4,000 WSI images daily (7 TB of data).
• It enables the comparison of scanner performance and replication of digital pathology image datasets.
• The facility supports the evaluation of clinical AI algorithms using standardized, multi-scanner data.

Conclusions:

• The AI FORGE facility is a novel resource for creating comprehensive and robust datasets for AI in pathology.
• Replicating slides across multiple WSI systems enhances AI generalizability and potentially reduces training data needs.
• This infrastructure is crucial for advancing AI applications in clinical tissue diagnosis.