Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

In AJR. American journal of roentgenology

OBJECTIVE. The purpose of our study was to develop a radiomics model based on preoperative MRI and clinical information for predicting recurrence-free survival (RFS) in patients with advanced high-grade serous ovarian carcinoma (HGSOC). MATERIALS AND METHODS. This retrospective study enrolled 117 patients with HGSOC, including 90 patients with recurrence and 27 without recurrence; 1046 radiomics features were extracted from T2-weighted images and contrast-enhanced T1-weighted images using a manual segmentation method. L1 regularization-based least absolute shrinkage and selection operator (LASSO) regression was performed to select features, and the synthetic minority oversampling technique (SMOTE) was used to balance our dataset. A support vector machine (SVM) classifier was used to build the classification model. To validate the performance of the proposed models, we applied a leave-one-out cross-validation method to train and test the classifier. Cox proportional hazards regression, Harrell concordance index (C-index), and Kaplan-Meier plots analysis were used to evaluate the associations between radiomics signatures and RFS. RESULTS. The fusion radiomics-based model yielded a significantly higher AUC value of 0.85 in evaluating RFS than the model using contrast-enhanced T1-weighted imaging features alone or T2-weighted imaging features alone (AUC = 0.79 and 0.74 and p = .02 and .01, respectively). Kaplan-Meier survival curves showed significant differences between high and low recurrence risk in patients with HGSOC by different models. The fusion model combining radiomics features and clinical information showed higher performance than the clinical model (C-index = 0.62 and 0.60, respectively). CONCLUSION. The proposed MRI-based radiomics signatures may provide a potential way to develop a prediction model and can help identify patients with advanced HGSOC who have a high risk of recurrence.

Li Hai Ming, Gong Jing, Li Rui Min, Xiao Ze Bin, Qiang Jin Wei, Peng Wei Jun, Gu Ya Jia

2021-Jul-14

MRI, machine learning, ovarian neoplasms, radiomics, recurrence