Using artificial intelligence to prioritize pathology samples: report of a test drive

Area of Science:

• Digital pathology
• Computational pathology
• Artificial intelligence in medicine

Background:

• Current pathology workflows face challenges addressable by automation.
• Digital transformation offers solutions for efficiency and accuracy in diagnostics.
• Artificial intelligence (AI) tools are emerging for complex pathological analysis.

Purpose of the Study:

• To evaluate Paige Pan Cancer, a novel AI tool for detecting invasive cancer.
• To assess the performance of this AI tool across 16 primary tissue types.
• To determine the tool's utility in clinical pathology practice.

Main Methods:

• Utilized Paige Pan Cancer, an AI tool based on the Virchow foundation model.
• Tested the tool on 62 haematoxylin and eosin-stained slide cases from the Ipatimup Pathology Laboratory.
• Evaluated performance on both tissue biopsies and resections.

Main Results:

• The AI tool achieved 93.3% sensitivity and 87.5% specificity in biopsies.
• In resections, sensitivity was 94.7% and specificity was 75.0%.
• Overall accuracy for cancer detection was 90.3%.

Conclusions:

• Paige Pan Cancer demonstrates high sensitivity as a multi-organ cancer screening tool.
• The AI tool shows promise for integration into clinical pathology practice.
• Further evaluation is warranted despite some observed misclassifications.