Frontiers in Oncology (Oct 2023)

Combined deep-learning MRI-based radiomic models for preoperative risk classification of endometrial endometrioid adenocarcinoma

  • Jin Yang,
  • Yuying Cao,
  • Fangzhu Zhou,
  • Chengyao Li,
  • Jiabei Lv,
  • Pu Li

DOI
https://doi.org/10.3389/fonc.2023.1231497
Journal volume & issue
Vol. 13

Abstract

Read online

BackgroundDifferences exist between high- and low-risk endometrial cancer (EC) in terms of whether lymph node dissection is performed. Factors such as tumor grade, myometrial invasion (MDI), and lymphovascular space invasion (LVSI) in the European Society for Medical Oncology (ESMO), European SocieTy for Radiotherapy & Oncology (ESTRO) and European Society of Gynaecological Oncology (ESGO) guidelines risk classification can often only be accurately assessed postoperatively. The aim of our study was to estimate the risk classification of patients with endometrial endometrioid adenocarcinoma before surgery and offer individualized treatment plans based on their risk classification.MethodsClinical information and last preoperative pelvic magnetic resonance imaging (MRI) of patients with postoperative pathologically determined endometrial endometrioid adenocarcinoma were collected retrospectively. The region of interest (ROI) was subsequently plotted in T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and diffusion-weighted imaging (DWI) MRI scans, and the traditional radiomics features and deep-learning image features were extracted. A final radiomics nomogram model integrating traditional radiomics features, deep learning image features, and clinical information was constructed to distinguish between low- and high-risk patients (based on the 2020 ESMO-ESGO-ESTRO guidelines). The efficacy of the model was evaluated in the training and validation sets of the model.ResultsWe finally included 168 patients from January 1, 2020 to July 29, 2021, of which 95 patients in 2021 were classified as the training set and 73 patients in 2020 were classified as the validation set. In the training set, the area under the curve (AUC) of the radiomics nomogram was 0.923 (95%CI: 0.865–0.980) and in the validation set, the AUC of the radiomics nomogram was 0.842 (95%CI: 0.762–0.923). The nomogram had better predictions than both the traditional radiomics model and the deep-learning radiomics model.ConclusionMRI-based radiomics models can be useful for preoperative risk classification of patients with endometrial endometrioid adenocarcinoma.

Keywords