Application of histogram analysis based on DCE-MRI in predicting metachronous liver metastasis and metastasis-free survival in rectal cancer.

BACKGROUND

This study aimed to construct histogram analysis (HA) based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) from primary tumor and mesorectum for early post-treatment risk stratification of metachronous liver metastasis (MLM) and prediction of metastasis-free survival (MFS) in rectal cancer.

METHODS

This retrospective study included preoperative images and clinical data of 251 patients between March 2019 and August 2023. Image segmentation was performed by manually delineating the primary tumor and mesorectum. The mean values of DCE-MRI perfusion parameters (K, K and V) and HA features (maximum, minimum, P10th, P50th, P90th, skewness, kurtosis, variance, and entropy) were compared between the two groups. The primary outcome measure was MFS, defined as the occurrence of MLM originating from rectal cancer or death from any cause after radical surgery. Multivariate Cox regression analysis and least absolute shrinkage and selection operator (LASSO) method were employed to screen features and construct nomogram of combined model that integrated clinicopathologic model, radiological model, and HA model from the tumor and mesorectum. The C-index, the time-independent area under the curve (AUC), and decision curve analysis (DCA) were performed to assess the clinical usefulness of nomogram.

RESULTS

The K entropy of tumor, V kurtosis of mesorectum, advanced T stage, lymphovascular invasion, MRI-defined mesorectal fascia invasion and lateral lymph node metastasis were demonstrated as independent prognostic factors of MFS. The combined model performed better (C-index = 0.895) than other models (C-index = 0.633, 0.702, and 0.823 respectively). Patients could be categorized into high-risk and low-risk groups with the nomogram (P < 0.001). The AUCs for nomogram predicting 1, 2, and 3-year MFS were 0.693, 0.833, and 0.921 respectively. The DCA also confirmed the larger clinical benefits of nomogram than other models.

CONCLUSIONS

The nomogram of combined model incorporating HA features, clinicopathologic characteristics, and radiological features was promising for early post-treatment risk stratification of MLM and prediction of the MFS in rectal cancer. HA features from the tumor and mesorectum could supply incremental value to guide postoperative individual follow-up plans.