Artificial Intelligence for Predicting Microsatellite Instability Based on Tumor Histomorphology: A Systematic Review

Area of Science:

• Oncology
• Computational Pathology
• Biomarker Discovery

Background:

• Microsatellite instability (MSI) and defective DNA mismatch repair (dMMR) are key biomarkers for immune checkpoint inhibitor (ICI) eligibility in advanced cancers.
• Current MSI/dMMR testing faces limitations due to high costs and resource constraints.
• Artificial intelligence (AI) offers a potential solution by predicting MSI/dMMR status from routine H&E stained histopathology slides.

Purpose of the Study:

• To systematically review the potential of AI-based histomorphological analysis in predicting MSI/dMMR status across various cancer types.
• To evaluate the performance and identify limitations of existing AI models for MSI/dMMR prediction.

Main Methods:

• A systematic literature search was conducted on PubMed and Embase for studies published up to September 2021.
• Included studies were analyzed for their design, reported performance metrics, and risk of bias.
• AI system performance was summarized for colorectal, gastric, and endometrial cancers.

Main Results:

• AI-based systems demonstrated excellent predictive performance for MSI/dMMR in colorectal cancer (highest AUC 0.972).
• Satisfactory performance was observed for gastric (highest AUC 0.81) and endometrial cancers (highest AUC 0.82).
• A significant risk of bias was identified in most of the reviewed studies.

Conclusions:

• AI holds substantial promise for predicting MSI/dMMR status, particularly in colorectal cancer, potentially streamlining biomarker assessment for ICI therapy.
• The findings suggest that AI-positive results might warrant confirmation, while AI-negative results may not require further testing.
• Further high-quality studies are needed to mitigate bias and validate AI models for broader clinical application.