Introduction

Microsatellite instability (MSI) is the condition of genetic hypermutability that results from impaired DNA mismatch repair (MMR). Many studies have demonstrated the role of high levels of microsatellite instability (MSI-H) as a prognostic marker and predictor of the response to chemotherapy in colorectal cancer (CRC), however, there was no non-invasive approach to preoperatively predict the MSI-H status in CRC patients. Recently, radiomics was introduced to extract high throughput data from anatomical images and advanced functional imaging (DWI, DCE, etc) and was correlated with clinic-pathological information (such as outcome and genomic data) to derive both diagnostic and prognostic imaging biomarkers. Therefore, we aimed to develop a multi-parametric MRI-based radiomics classifier to non-invasively and preoperatively predict the MSI-H status in CRC.

Methods

This study retrospectively included 232 CRC patients (MSI-H CRC, n = 85; non-MSI-H, n = 147). A total of 2368 radiomics features were extracted from T1WI, T2WI, DWI and T1-contrast enhanced images of the CRC lesions. Data dimension reduction was conducted by the minimum redundancy maximum relevance (mRMR) algorithm, then radiomics features were selected using the least absolute shrinkage and selection operator method (LASSO) logistic regression model, with which a radiomics classifier was built. Association between radiomics classifier and CRC MSI-H status was investigated and the classification performance of the radiomics classifier was explored with the receiver operating characteristics (ROC) curve analysis.

Results and discussion

A multi-parametric MRI-based radiomics classifier with 21 radiomics features was successfully built, and it was an independent predictor for determining the MSI-H status in CRC patients, which could divide the CRC patients into MSI-H and non-MSI-H status (p<0.01). The median score of the radiomics classifier in MSI-H was higher than those of non-MSI-H. In terms of the classification performance of the radiomics classifier in MSI-H status of CRC patients, the C-index was 0.823 (95%CI: 0.794-0.847) with a corrected C-index of 0.811 through internal validation by using bootstrapping validation. The predicting model yielded a sensitivity of 0.893 and specificity of 0.825.

Conclusion

This study presented a multi-parametric radiomics classifier for preoperative and individualized prediction of MSI-H status in CRC patients. The potential application of this radiomics approach may aid the prognostic evaluation and decision-making in CRC patients.

Acknowledgements

No acknowledgement found.

References

No reference found.