Background Multiple studies have demonstrated the predictive value of machine learning-based radiomic approaches to CT imaging in colorectal cancer microsatellite instability (MSI) status diagnosis. This study aimed to identify and formulate potential CT-based imaging characteristics available to the general radiologist without access to radiomics, to differentiate MSI status amongst colorectal cancer tumors.

Methods 109 patients with an average age of 61 years (58 male, 51 female) who were diagnosed with MSI-positive colorectal adenocarcinomas between 2011 and 2018 and received imaging at Banner MD Anderson Cancer Center. This was a retrospective analysis where various imaging features of the primary mass and metastasis were analyzed, including location, size, attenuation (compared to the liver and metastases, if present), growth pattern, tumor margin, primary mass area, and mesenteric infiltration. This data was compared with tumor biopsy MSI staging to identify correlations. Patient information (age, sex, tumor staging) was obtained from medical records. Statistical analysis was performed using chi-square, t-test, and Mann-Whitney U tests.

Results Five tumor characteristics showed a statistically significant relationship with MSI status. Clinical staging differed significantly between MSI groups (p=0.012). CRC tumors with distant metastasis are more likely to be MSI-L, and those with regional metastasis are more likely to be MSI-H (p<0.001). Primary mass area was larger in the MSI-H group compared to the MSI-L group (p<0.001). CRC tumors in the left colon are more likely to be MSI-L, while those in the right colon are more likely to be MSI-H (p<0.001). MSI-H samples had a lower primary tumor Hounsfield Units standard deviation (SD) than MSI-L samples (p=0.002). Mesenteric infiltration showed no significant difference (p=0.189). Both MSI-L and MSI-H were likely to be well-defined, with no statistical difference in definition (p=1.000). They were also likely to be homogenous, with no statistical difference in homogeneity (p=0.498). Both MSI groups were more likely to exhibit mucosal rather than mass-like growth, with no statistical difference in growth type (p=0.127). The distribution of Mean Primary Tumor Hounsfield Unit: hyperdense region and mean Liver Hounsfield Unit was the same across MSI categories (p=0.162 and p=0.105, respectively). Both MSI-L and MSI-H were more likely to be hypoattenuating, with no significant relationship between tumor enhancement and MSI status (p=0.376).

Conclusion Multiple CT imaging characteristics were relevant in predicting the classification of MSI in CRC tumors, holding potential clinical relevance in guiding individualized therapy. While these results were statistically significant, more research must be done to validate these findings.