Associations between MRI radiomic phenotypes and clinical outcomes in endometrial cancer: Implications for preoperative risk stratification.

OBJECTIVES

This study aimed to investigate the correlation between imaging phenotypes of endometrial cancer (EC) and clinical, pathologic, and molecular characteristics, as well as disease-free survival (DFS).

METHODS

The clinical, pathologic, and molecular characteristics, along with MRI radiomics features, of 356 patients with EC were collected retrospectively. The patients were divided into 2 groups based on radiomics features using unsupervised machine learning. The obtained characteristics and DFS of patients were compared between the various imaging phenotypes.

RESULTS

The lesions with deep myometrial invasion (DMI), lymphovascular space invasion (LVSI), cervical stromal invasion (CSI), lymph node metastasis, aggressive histologic type, advanced postoperative International Federation of Gynecology and Obstetrics (FIGO) stage, overexpression of p53, and absent expression of estrogen receptor or progesterone receptor were associated with poor DFS. Two clusters were identified and defined as imaging phenotype 1 and 2, respectively. Compared with phenotype 2, phenotype 1 exhibited a higher correlation with DMI (33.7 % vs 13.0 %), LVSI (23.8 % vs 9.2 %), CSI (16.3 % vs 3.8 %), aggressive histologic type (36.0 % vs 17.4 %), and advanced FIGO stage (IB or higher, 43.6 % vs 22.3 %) (p < 0.001). The incidence of p53 overexpression was higher in phenotype 1 than in phenotype 2 (20.2 % vs 8.5 %, p = 0.022). Survival analysis exhibited a higher risk of poor DFS in phenotype 1 than in phenotype 2 (log-rank p = 0.002).

CONCLUSION

EC imaging phenotypes identified through MRI radiomics features were associated with pathologic, molecular characteristics, and DFS, suggesting potential for preoperative risk stratification.