Preoperative Assessment of MRI‐Invisible Early‐Stage Endometrial Cancer With MRI‐Based Radiomics Analysis

Abstract

BackgroundRadiomics‐based analyses have demonstrated impact on studies of endometrial cancer (EC). However, there have been no radiomics studies investigating preoperative assessment of MRI‐invisible EC to date.PurposeTo develop and validate radiomics models based on sagittal T2‐weighted images (T2WI) and T1‐weighted contrast‐enhanced images (T1CE) for the preoperative assessment of MRI‐invisible early‐stage EC and myometrial invasion (MI).Study TypeRetrospective.PopulationOne hundred fifty‐eight consecutive patients (mean age 50.7 years) with MRI‐invisible endometrial lesions were enrolled from June 2016 to March 2022 and randomly divided into the training (n = 110) and validation cohort (n = 48) using a ratio of 7:3.Field Strength/Sequence3‐T, T2WI, and T1CE sequences, turbo spin echo.AssessmentTwo radiologists performed image segmentation and extracted features. Endometrial lesions were histopathologically classified as benign, dysplasia, and EC with or without MI. In the training cohort, 28 and 20 radiomics features were selected to build Model 1 and Model 2, respectively, generating rad‐score 1 (RS1) and rad‐score 2 (RS2) for evaluating MRI‐invisible EC and MI.Statistical TestsThe least absolute shrinkage and selection operator logistic regression method was used to select radiomics features. Mann–Whitney U tests and Chi‐square test were used to analyze continuous and categorical variables. Receiver operating characteristic curve (ROC) and decision curve analysis were used for performance evaluation. The area under the ROC curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value were calculated. A P‐value <0.05 was considered statistically significant.ResultsModel 1 had good performance for preoperative detecting of MRI‐invisible early‐stage EC in the training and validation cohorts (AUC: 0.873 and 0.918). In addition, Model 2 had good performance in assessment of MI of MRI‐invisible endometrial lesions in the training and validation cohorts (AUC: 0.854 and 0.834).Data ConclusionMRI‐based radiomics models may provide good performance for detecting MRI‐invisible EC and MI.Evidence Level3Technical EfficacyStage 2

Development and validation of a combinatorial MRI sequence-based radiomics model for preoperative prediction of microsatellite instability status in rectal cancer

“…In the present study, the combination of T2WI and T1CE sequences showed the highest prediction performance among all combinations of sequences. This was consistent with the results reported by Jiang et al [23] , who found that a radiomics model using T2WI and T1CE sequences provided good performance for detecting MRI-invisible early-stage endometrial cancer and myometrial invasions. Hu et al [24] combined T1CE and T2WI radiomics analysis of cervical mucosa to predict early cervical cancer not visible by MRI.…”

Advances in Radiomics Research for Endometrial Cancer: A Comprehensive Review

“…In a parallel development, Yan, Ma et al [20] developed a nomogram for predicting ovarian preservation in early-stage EC, achieving an AUROC of 0.96, thereby surpassing the performance of radiologists (0.80 and 0.86). To address the challenge of detecting non-visible EC in MRI, Jiang's team [21] developed two radiomics models for distinguishing non-visible EC and MI, both demonstrating excellent performance with average AUROCs of 0.896 and 0.844, respectively. The current body of research indicates that classifiers based on radiomics features, either standalone or in conjunction with selected clinical and traditional imaging features, hold high diagnostic efficacy, potentially matching or exceeding the capabilities of seasoned radiologists, thereby enhancing diagnostic precision.…”

Editorial for “Preoperative Assessment of MRI‐Invisible Early‐Stage Endometrial Cancer With MRI‐Based Radiomics Analysis”

“…4 In this issue of the Journal of Magnetic Resonance Imaging, Chen et al apply radiomic analysis to patients suspected of having endometrial cancer without visible lesions at MR imaging. 5 They examined a total of 2074 imaging features in 130 patients with MRI-invisible suspected endometrial pathology. Threedimensional regions of interest were placed on T2-weighted images and T1-weighted images obtained after gadolinium contrast.…”