H&E-based MSI/MMR testing with AI in colorectal cancer: a multi-centred blinded evaluation

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Mismatch repair (MMR) deficiency is found in 10-20% of colorectal cancer (CRC) cases, causing microsatellite instability (MSI).
• Current MSI/MMR testing is crucial for CRC management but faces limitations due to high costs and long turnaround times.
• There is a need for accessible and cost-effective diagnostic alternatives for MSI/MMR status in CRC.

Purpose of the Study:

• To validate the performance of PANProfiler Colorectal (PPC), an AI-based biomarker test, for determining MSI/MMR status.
• To assess PPC's accuracy using routine haematoxylin and eosin (H&E)-stained slides in a multi-centre setting.
• To evaluate PPC as a potential rapid and scalable alternative to conventional MSI/MMR diagnostic methods.

Main Methods:

• A blinded, multi-centre validation study was conducted involving 3,576 whole slide images from 1,243 CRC patients.
• PANProfiler Colorectal (PPC), an AI algorithm, analyzed H&E-stained slides to determine MSI/MMR status.
• PPC results were compared against standard MSI/MMR testing methods.

Main Results:

• PPC successfully generated definitive results for 86.55% of the analyzed slides.
• The AI test achieved high overall agreement (93.83%) with standard testing.
• Positive agreement was 92.54% and negative agreement was 94.02%, demonstrating strong concordance.

Conclusions:

• PANProfiler Colorectal (PPC) accurately determines MSI/MMR status directly from routine H&E-stained colorectal cancer slides.
• The AI-based test offers a rapid, scalable, and cost-effective alternative to current diagnostic approaches.
• PPC has the potential to improve the accessibility and clinical utility of MSI/MMR testing in CRC management.