Deep learning for the detection of microsatellite instability from histology images in colorectal cancer: A systematic literature review

Abstract

Microsatellite instability (MSI) or deficient mismatch repair (dMMR) is a clinically important genetic feature affecting 10–15% of colorectal cancer (CRC) patients. Patients with metastatic MSI/dMMR CRC are eligible for therapy with immune checkpoint inhibitors, making MSI/dMMR the most important immuno-oncological biomarker in CRC. Gold standard tests for detection of MSI/dMMR in CRC are based on wet laboratory tests such as immunohistochemistry (IHC) or DNA extraction with subsequent polymerase chain reaction (PCR). However, since 2019, advances in Deep Learning (DL), an Artificial Intelligence (AI) technology, have enabled the prediction of MSI/dMMR directly from digitized routine haematoxylin and eosin (H&E) histopathology slides with high accuracy. In addition to the initial proof-of-concept publication in 2019, twelve subsequent studies have refined, improved, and further validated this approach. At this moment, MSI/dMMR prediction using Deep Learning has become a widely used benchmark task for academic studies in the field of computational pathology. Beyond academic use, this assay has attracted commercial interest from companies with the possibility of approval as a diagnostic device in the near future. In this review, we summarize and quantitatively compare the existing evidence on Deep-Learning-based detection of MSI/dMMR in CRC and discuss the need for further improvement and potential for integration into routine pathological workflows. Ultimately, this DL-based method could facilitate the identification of patients eligible for treatment with immune checkpoint inhibitors by pre-screening or replacement of current methods.