AI-Enhanced H&E-Based Assessment of MSI and MMR in Colorectal Cancer
Overview
PANProfiler Colorectal (PPC), an AI-based test analyzing H&E-stained slides, demonstrated high accuracy in determining MSI/MMR status in colorectal cancer across three UK centers. The blinded multi-center validation showed PPC achieved over 93% overall agreement with standard testing methods, providing a rapid and scalable alternative.
Background
Mismatch repair (MMR) deficiency occurs in 10–20% of colorectal cancers (CRC), leading to microsatellite instability (MSI), which has important diagnostic, prognostic, and therapeutic implications. Current MSI/MMR testing methods, including immunohistochemistry (IHC) and PCR, are limited by cost, turnaround time, and tissue requirements. Morphological features associated with MSI/dMMR on routine H&E-stained slides suggest potential for AI-based detection. PPC leverages these features to provide MSI/MMR status directly from H&E slides, aiming to improve accessibility and efficiency of testing.
Data Highlights
Metric
Value
Number of whole slide images
3,576
Number of CRC patients
1,243
Definitive result rate
86.55%
Overall percent agreement (OPA)
93.83%
Positive percent agreement (PPA)
92.54%
Negative percent agreement (NPA)
94.02%
C-statistics
>0.92 across cohorts
Indeterminate result rate
13.45%
Key Findings
PPC achieved high concordance with standard MSI/MMR testing, with overall agreement of 93.83%.
The AI model produced definitive results in 86.55% of H&E-stained slides, indicating robust applicability.
Positive and negative percent agreements were 92.54% and 94.02%, respectively, demonstrating balanced sensitivity and specificity.
Performance was consistent across three independent UK cohorts, supporting generalisability.
Indeterminate results occurred in only 13.45% of cases, reflecting a low test replacement rate.
PPC can analyze routine H&E slides without requiring additional tissue or specialized infrastructure.
Clinical Implications
PPC offers a rapid, cost-effective, and scalable alternative to conventional MSI/MMR testing methods, potentially increasing testing rates and reducing turnaround times. Its ability to utilize standard H&E slides may alleviate tissue and resource constraints, facilitating broader implementation in routine colorectal cancer diagnostics. This AI-driven approach could streamline patient stratification for immunotherapy and Lynch syndrome screening.
Conclusion
The blinded multi-center study validates PPC as a reliable AI-based tool for MSI/MMR status determination from H&E-stained slides, with performance comparable to standard methods. PPC holds promise to enhance accessibility and efficiency of MSI/MMR testing in colorectal cancer care.
by Cher Bass, Foivos Ntelemis, Julian Schmidt, Steffen Wolf, André Geraldes, Debapriya Mehrotra, Shikha Singhal, Narender Kumar, Angelica Marcia, Nicholas Bennett, Oscar Maiques, Mitchell Hyde, Bejal Mistry, Grace Rogerson, Michele Cummings, Clare Freer, Elizabeth Walsh, Manuel Salto-Tellez, Maurice Loughrey, In Hwa Um, David J. Harrison, Richard Clarkson, James Blackwood, J. Carl Barrett, Jakob Nikolas Kather, Nicolas M. Orsi, Pahini Pandya, Salim Arslan
