Colorectal cancers (CRCs) can vary considerably from patient to patient, making histologic interpretation complex. To aid in interpreting the large amount of data available on hematoxylin and eosin (H&E) slides, researchers have developed QuantCRC, a quantitative segmentation algorithm that can assess known histologic features in digitized CRC images.¹ This information can then be used in predicting which patients are likely to experience a future recurrence.

For this analysis, 6468 digitized H&E slides were obtained from 8 sites within the Colon Cancer Family Registry (CCRF), which included locations in Australia, Canada, and the United States. QuantCRC was used to quantify these 15 parameters of the tumor and tumor environment: percentages of tumor and stroma cells; tumor to stroma ratio; percentages of tumor budding/poorly differentiated clusters and mucin within the tumor; percentage of necrosis within the tumor bed; percentage high-grade cancerous cells; percentage of signet ring cell carcinoma; number of tumor-infiltrating lymphocytes per mm² of tumor; percentages of immature, mature, and inflammatory stroma cells within the tumor bed; and percentages of immature, mature, and inflammatory cells within the stromal region. (The accuracy of QuantCRC has been shown in previous research.²) These data were combined with TNM stage and mismatch repair status to train a prognostic model with information from 1928 tumors. The model was validated on data from another 1421 tumors and then used to predict recurrence-free survival (RFS).

The resulting predicted median recurrence rates at 36 months were 32.7% for high-risk stage III CRC and 13.4% for low-risk stage III, after adjusting for established risk factors. For stage II disease, the predicted 36-month recurrence rates were 15.8% for high risk and 5.4% for low risk.

According to the authors, these results show that quantitative digital pathology may provide another tool—in addition to routine pathologic reporting—to help guide therapy and surveillance in CRC. This information can allow clinicians to classify tumors into relevant prognostic groups using standardized results, reducing the variability that may occur with manual interpretation. The limitations, however, include the need for digitizing the slides and the use of commercial software.

References

1. Pai RK, Banerjee I, Shivji S, et al. Quantitative pathologic analysis of digitized images of colorectal carcinoma improves prediction of recurrence-free survival. Gastroenterology. 2022;163(6):1531-1546.e8. doi:10.1053/j.gastro.2022.08.025
2. Pai RK, Hartman D, Schaeffer DF, et al. Development and initial validation of a deep learning algorithm to quantify histological features in colorectal carcinoma including tumour budding/poorly differentiated clusters. Histopathology. 2021;79(3):391-405. doi:https://doi.org/10.1111/his.14353