104P - The value of functional MR-imaging signature model for early prediction of chemotherapy response and its guidance for regimen adjustment to improve efficacy

Background

We evaluated the feasibility of functional magnetic resonance (MR) signature in predicting clinical response to chemotherapy in colorectal liver metastatic (CLM) patients.

Methods

From August 2016 to January 2023, eligible CLM patients were enrolled and functional MR was performed at baseline and one cycle after chemotherapy. The diffusion kurtosis radiomic texture features were extracted and functional MR-imaging signature model was built by the R package called “glmnet” to predict the efficacy of treatment. The initial 100 cases were utilized as training set, the following 48 cases as validation set, and the latter 48 cases as intervention validation set. The primary endpoint was the accuracy of MR-predicted response (ORR) of liver metastases.

Results

The functional MR signature was established and showed a good performance of response prediction (AUC was 0.818 in training cohort and 0.755 in validation cohort). In training set, ORRs were 9.5% (4/42) and 72.4% (42/58) in high-risk and low-risk subgroups. In validation set, the ORRs were 23.81% (5/21) and 74.07% (20/27) in high-risk and low-risk subgroups, respectively (P=0.002). Worse PFS and OS were observed in high-risk population in these two sets. In intervention set, chemotherapy regimen was changed in 22.9% (11/48) patients who were predicted as high-risk by the model, and the ORR reached 54.6 % (6/11), which was higher than that in high-risk subgroups in training and validation set.

Conclusions

Functional MR signature effectively predicts chemotherapy response and long-term survival of patients with CLM, and regimen adjustment guided by the model significantly improve the ORR.

Clinical trial identification

NCT03088163.

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.