MRI-based radiomics and ADC values are related to recurrence of endometrial carcinoma: a preliminary analysis

Endometrial carcinoma (EC) is the sixth most commonly diagnosed cancer in women. With the increasing prevalence of risk factors such as physical inactivity and overweight, the incidence of endometrial cancer continues to rise [1, 2]. Surgery is the main method for the initial treatment of EC including laparoscopic or robotic resection of uterus, cervix, fallopian tube and ovary, and sentinel lymph node assessment [3]. Although most endometrial cancers can be diagnosed at an early stage and have 5-year survival rate of over 90%, recurrence and final mortality approximately occur in 20% of endometrioid carcinoma (type I) and 50% of non-endometrioid carcinoma (type II) [4‐6]. Women with recurrent or metastatic diseases have 5-year survival rates as low as 17–55% [7, 8]. Unfortunately, little progress has been made in improving the survival rate of EC in the past decades. Therefore, early identification of risk factors for recurrence is an important challenge to improve the prognosis of EC patients.

Clinicopathological factors, such as FIGO stage, pathological type and muscular invasion, etc. are common prognostic factors that affect the formulation of surgical plan, but they can only be accurately evaluated after surgery [9]. Due to the limitations of specimen collection, preoperative endometrial biopsy may be difficult to fully reflect the characteristics and heterogeneity of the tumor. Therefore, it is necessary to develop an early, comprehensive and non-invasive evaluation method to evaluate the possible adverse prognosis of EC. MRI is an important tool to preliminarily assess the extent of EC lesions. The apparent diffusion coefficient (ADC) of diffusion-weighted imaging (DWI) can reflect the malignancy of the tumor, which has been proved to be valuable in the diagnosis, typing and grade of EC [10‐12]. Gharibvand et al. [11]. reported that DWI sequence (ADC value) had high accuracy in diagnosing benign endometrial lesions and differentiating malignant lesions. Yashar et al. [12]. conducted a mate-analysis involving 11 studies and found that ADC values had good diagnostic accuracy in differentiating EC from benign lesions, with combined sensitivity and specificity of 93%(87~96%; I² = 41.19%) and 94% (88% ~ 97%; I² = 46.91%). In addition, jiang et al.’s study [13] showed that ADC value of grade 1 EC patients was significantly higher than that of grade 3 patients, and ADC value of patients with high expression of Ki-67 was significantly lower than that of patients with low expression of Ki-67. These studies suggest that ADC value has the potential to predict EC biological behavior, but its role in the prognostic assessment of EC remains unclear.

Radiomics mines pixel distribution features from radiological images and transforms them into quantitative data, reflecting the heterogeneity within tumors [14]. Radiomic parameters derived from MRI, CT and PET-CT have been suggested as effective tools for diagnosis, risk assessment or treatment response of malignant tumors [15‐17]. The nomogram of radiomics extracted from ADC sequence can be used to predict EC histological classification before operation [18]. Rodríguez-Ortega et al. [19] combined texture parameters extracted from T2 weighted images (T2WI) and ADC sequences with semi-quantitative parameters of DWI and dynamic contrast-enhanced (DCE) for preoperative evaluation of myometrial invasion. A multicenter study established a model for evaluating pelvic lymph node status based on radiomics of MRI, which helped radiologists improve diagnostic efficiency [20]. In addition, Wang et al. [21] found that the radiomic feature of GLCMEntropy extracted from PET/CT for tumors is a potential predictor of PD1 expression in EC. The purpose of this study is to explore whether postoperative recurrence of patients with EC can be reflected in MRI-based ADC value and radiomics.

Traditionally, prognostic factors for EC include surgical stage, histological factors [22, 23], and molecular factors such as P53 mutation [24]. This information is often only available from histological evaluation after surgery, so preoperative risk assessment of recurrence is limited. As a routine preoperative imaging examination of EC, MRI can provide anatomical information, and DWI and radiomics analysis may reveal the prognostic information of EC. In addition to clinical and pathological factors, we also investigated the correlation between ADC values (including ADC_mean, ADC_min, ADC_max) and radscore and recurrence in this study. We identified risk factors associated with recurrence and reduced survival of EC after surgery, including FIGO stage, pathologic type, depth of muscular invasion, ADC_mean, ADC_min and radscore. Non-endometrioid cancer, ADC_mean and radscore were independent predictors of EC recurrence, which can be used as clinically relevant tumor markers in preoperative risk stratification and prognosis assessment of EC.

Although the number of recurrent patients is small in our study, we identified the clinical and pathological risk factors related to recurrence of EC patients including FIGO stage, pathological type, and myometrial invasion depth. These factors have been associated with recurrence or poor outcome in EC in other studies. Women who were initially diagnosed with advanced disease (FIGO III- IV) had a higher risk of recurrence and were more likely to develop extrapervic metastases [8]. Patients with low-grade (grade 1-2) endometrioid cancer (type I) often have better outcomes than those with high-grade (grade 3) endometrioid cancer and non-endometrioid cancer (type II) [25]. A study exploring recurrence factors for stage I endometrioid adenocarcinoma observed that large tumor size, and muscular invasion were the most important predictors of recurrence [9]. The study of Bosse et al. [22] confirmed that substantial LVSI was the strongest prognostic factor for recurrence and metastasis and overall survival of EC.

Notably, we did not find age of onset, tumor size, LVSI or LNM as factors related to the increased risk of recurrence. The subjects of our study were concentrated in the 50-60 years old, which was a homogeneous population, with few patients under 50 years old or over 70 years old. In our study, the tumor size was defined as the maximum diameter of the tumor, which was often the extent to which tumor tissue invaded the endometrium rather than the depth. So, there was no significant difference in tumor size between the relapsed and non-relapsed groups, and tumor size had no significant effect on survival. Less cases with positive lymph node metastasis and LVSI may be the reason why LNM and LSVI had no effect on survival time.

In recent years, many studies have demonstrated the value of MRI in predicting histopathological features of gynecological tumors. Satta et al. [26] prospectively analyzed the association between preoperative MRI imaging parameters and histopathology in 44 EC patients, and found that the quantitative parameters of preoperative DWI, intravoxel incoherent motion (IVIM) and dynamic contrast-enhanced (DCE) MRI can reflect the physiological and microscopic characteristics of EC. Chen et al. [10] reported that type II EC had lower ADC values, and ADC values were important when identifying type II and type I ECs. The study of Jiang et al. [13] indicated that the ADC values of high grade, stage IB and high Ki-67 expression patients were significantly lower than those of low grade, stage IA and low Ki-67 expression patients with EC. Zhang et al. [27] reported that lower ADC were observed in tumor with deep myometrial invasion and LVSI than tumor without deep myometrial invasion and LVSI. Similarly, our previous studies [28] explored the role of ADC values in EC histological features and demonstrated that lower ADC values were associated with higher grade, non- endometrioid subtype and a higher risk of deep muscular-layer invasion and lymph node metastasis. These findings suggest a potential association between ADC related parameters and adverse outcomes in EC. Our present study found that lower ADC_mean and ADC_min value were related to the recurrence of EC and ADC_mean was shown to be an independent predictor of DFS in multivariate analysis. In addition, we also found that ADC_min was significantly negatively correlated with Ki-67 expression level. Many studies have explored the correlation between ADC related parameters and histological molecules in different tumors. Ma et al. [29] reported that the Ki-67 PI, HIF-1α expression and VEGF expression in prostate cancer were correlated inversely with ADC. Surov et al. [30] conducted a multicenter study of breast cancer and found that ADC values were significantly correlated with Ki-67 expression in breast cancer. Similarly, Zhang et al. [31] reported that lower mean ADC values were associated with higher Ki-67 expression in Type I ovarian epithelial cancer. ADC values can reflect the diffusion degree of water molecules and is closely related to cell density. Yan et al. [20] and Reyes-Pérez et al. [32] explained that the decrease of ADC can act as markers for tumor’s high cellularity, proliferation, perfusion and less extracellular space, which means higher tumor load, tumor residue and recurrence. Therefore, DWI and ADC values may be noninvasive techniques for evaluating endometrial cancer cell proliferation So, it is not difficult to explain the correlation between ADC and prognosis and recurrence.

Heterogeneity is an important characteristic of malignant tumors and the basis of tumor recurrence and metastasis [33, 34]. Radiomics can extract features that cannot be observed by naked eyes and convert them into data information to quantify the heterogeneity in the image [35, 36]. Radiomics derived from MRI have been proposed as a reliable tool for accurate diagnosis and risk assessment in several cancer types, e.g. in cervical [37], brain [38], and breast [39]. Similarly, a study has reported that radiomics analysis based on MRI was an effective tool for preoperative risk stratification in EC [40]. In this context, MRI-based radiomics analysis was available to predict recurrence in patients with EC, and 2 radiomic parameters were selected to predict recurrence. Among these, Glcm_DifferenceEntropy was a measure of the randomness/variability in neighborhood intensity value differences, lbp features were the observation of the whole tumor range and the extraction of high-dimensional features. The radscore was calculated according to the combination of regression coefficients of each parameter, which was an independent predictor of relapse in both univariate and multivariate analyses. Patients were grouped into risk groups based on radscore, and patients with higher radscore had worse DFS. Sigmund et al. [41]. reported that thirteen MRI-derived tumor radiomic parameters significantly predicted reduced recurrence- and progression-free survival in univariable Cox regression analysis, and T1c_Kurtosis2 was the top-ranked prognostic texture parameter independently predicted reduced survival. The study of Yan et al. [16] manifested that MRI-based radiomics achieved high diagnostic performance for predicting LVSI of EC preoperatively, and was helpful for early identification of poor prognosis. In addition, the whole-tumor radiomic features were found to significantly predict progression-free survival at hazard ratios of 4.6–9.8 in the research of Fasmer et al. [42], albeit in a small sample size. Although the sample size, MRI sequence, and radiomic parameter extraction methods were different in the previous study, we cannot deny that radiomics and texture analysis may mine more prognostic information than clinical factors, and can be used as a biomarker to assist clinical practice.

Our current study expanded the sample size compared with other previous studies and simultaneously explored the correlation between clinicopathological factors, DWI quantitative parameters, and radiomics and EC recurrence. There are some limitations in our research. This is a single-center study, which needs to be further verified by a large multi-center database. The absence of stratification of staging and pathologic types may influence the estimation of survival and recurrence outcomes. A short follow-up time and less recurrence cases may limit the predictive value of the covariates.

In conclusion, EC patients are at certain risk of recurrence, and early identification of risk factors for recurrence and enhanced treatment intensity are significant to improve prognosis. ADC_mean values and MRI-derived radiomics may provide additional prognostic information in addition to traditional prognostic factors.