AI for pathologists: a universal lymph node metastasis detection app that enhances efficiency while preserving...

Area of Science:

• Digital pathology
• Artificial intelligence in medicine
• Cancer diagnostics

Background:

• Pathologist workload and shortages contribute to diagnostic errors.
• Manual tasks like mitosis counting and metastasis detection are time-consuming and unreliable.
• Digital pathology and AI offer potential solutions for improving efficiency and accuracy.

Purpose of the Study:

• To evaluate the generalizability of an AI application for lymph node metastasis detection across various cancer types and organs.
• To assess the impact of AI assistance on pathologist efficiency and diagnostic accuracy.

Main Methods:

• A commercial AI tool (Visiopharm app), initially trained on breast and colon cancer lymph node metastases, was used.
• The AI tool analyzed 172 slides from 12 cancer types across 15 organ systems.
• Pathologist time spent searching for metastasis was measured with and without AI assistance.

Main Results:

• The AI application successfully detected lymph node metastases from 12 distinct cancer types across 15 organ systems.
• AI assistance reduced the average time pathologists spent searching for metastasis (54.7s to 42.1s per slide).
• Diagnostic accuracy was maintained when using AI-generated annotations.

Conclusions:

• AI applications trained on specific datasets can generalize to detect metastases from multiple cancer types and organs.
• AI tools can significantly reduce pathologist workload and improve efficiency without compromising accuracy.
• Integrating AI into digital pathology workflows can help address pathologist shortages and enhance patient care.