Myeloid response evaluated by noninvasive CT imaging predicts post-surgical survival and immune checkpoint therapy benefits in patients with hepatocellular carcinoma

Abstract

Background: The potential of preoperative CT in the assessment of myeloid immune response and its application in predicting prognosis and immune-checkpoint therapy outcomes in hepatocellular carcinoma (HCC) has not been explored.

Methods: A total of 165 patients with pathological slides and multi-phase CT images were included to develop a radiomics signature for predicting the imaging-based myeloid response score (iMRS). Overall survival (OS) and recurrence-free survival (RFS) were assessed according to the iMRS risk group and validated in a surgical resection cohort (n = 98). The complementary advantage of iMRS incorporating significant clinicopathologic factors was investigated by the Cox proportional hazards analysis. Additionally, the iMRS in inferring the benefits of immune checkpoint therapy was explored in an immunotherapy cohort (n = 36).

Results: We showed that AUCs of the optimal radiomics signature for iMRS were 0.941 [95% confidence interval (CI), 0.909-0.973] and 0.833 (0.798-0.868) in the training and test cohorts, respectively. High iMRS was associated with poor RFS and OS. The prognostic performance of the Clinical-iMRS nomogram was better than that of a single parameter (p < 0.05), with a 1-, 3-, and 5-year C-index for RFS of 0.729, 0.709, and 0.713 in the training, test, and surgical resection cohorts, respectively. A high iMRS score predicted a higher proportion of objective response (vs. progressive disease or stable disease; odds ratio, 2.311; 95% CI, 1.144-4.672; p = 0.020; AUC, 0.718) in patients treated with anti-PD-1 and PD-L1.

Conclusions: iMRS may provide a promising method for predicting local myeloid immune responses in HCC patients, inferring postsurgical prognosis, and evaluating benefits of immune checkpoint therapy.

Keywords: hepatocellular carcinoma; immunotherapy; myeloid cells; prognosis; radiomics.

Figure 1

Study design. The training cohort and test cohort that contain data from patients with CT images and pathology slides were used to develop the radiomics signature of MRS. Two additional cohorts were used to validate the clinical and prognostic value of this radiomics signature. The surgical resection cohort comprised clinical data and the corresponding imaging data from patients with surgical resection. The immunotherapy cohort comprised advanced HCC patients who had been treated with anti-PD-1 or anti-PD-L1 therapy. MRS, myeloid response score.

Figure 2

Radiomics workflow. (A) Representative imaging and pathology data of patients from MRS-high and MRS-low groups. (B) Radiomics workflow. MRS-PM represents the mp-CT radiomics signature with the best predictive performance for MRS. Prognosis-PM represents the prognostic Clinical-iMRS nomogram by combining iMRS and the significant clinicopathological features. NP, non-contrast phase; AP, arterial phase; PVP, portal venous phase; ICC, intraclass correlation coefficient; PCC, Pearson correlation coefficient; RFE, recursive feature elimination; iMRS, imaging-based MRS; mp-CT, multi-phase CT.

Figure 3

Performance of the radiomics signature. (A) ROC curves of the optimal mp-CT radiomics signature in cohorts 1 and 2. (B) Decision curve analysis. (C, D) iMRS of patients in pMRS-high or -low groups in cohort 1 (C) and cohort 2 (D). Mann–Whitney test, ***p < 0.001. (E, F) The correlations between the number of tumor-infiltrating CD11b⁺ cells and iMRS in cohort 1 (E) and cohort 2 (F). Spearman correlation analysis. ROC, receiver operator characteristic; pMRS, pathological MRS.

Figure 4

Prognostic and clinical value of the iMRS. (A–F) Overall survival and recurrence-free survival of patients relative to iMRS in cohort 1, cohort 2, and the surgical resection cohort. (G) iMRS of patients with complete CR/PR or PD/SD to anti-PD-1/PD-L1 therapy in the immunotherapy cohort. Mann–Whitney test, *p < 0.05. (H) Overall survival of patients relative to iMRS in immunotherapy cohort. CR, complete response; PR, partial response; PD, progressive disease; SD, stable disease.

Figure 5

Clinical-iMRS nomogram and predictive performance evaluation. (A) The Clinical-iMRS nomogram to predict recurrence-free survival for post-surgical HCC patients. (B–D) Calibration curves for the nomogram in cohort 1 (B), cohort 2 (C), and the surgical resection cohort (D). (E–G) C-index for assessing the prognostic value of the clinical score, iMRS, and combined Clinical-iMRS score in cohort 1 (E), cohort 2 (F), and the surgical resection cohort (G). C-index, Harrell concordance index.

Figure 6

Prognostic value of the combined Clinical-iMRS model. (A–F) Overall survival and recurrence-free survival of patients relative to the Clinical-iMRS model (high risk or low risk) in cohort 1, cohort 2, and the surgical resection cohort.