Developing and validating ultrasound-based radiomics models for predicting high-risk endometrial cancer.

OBJECTIVES: The primary aim of this study was to develop and validate radiomics models, applied to ultrasound images, capable of differentiating from other cancers high-risk endometrial cancer, as defined jointly by the European Society for Medical Oncology, European Society of Gynaecological Oncology and European Society for Radiotherapy & Oncology (ESMO-ESGO-ESTRO) in 2016. The secondary aim was to develop and validate radiomics models for differentiating low-risk endometrial cancer from other endometrial cancers. METHODS: This was a multicenter, retrospective, observational study. From two participating centers, we identified consecutive patients with histologically confirmed diagnosis of endometrial cancer who had undergone preoperative ultrasound examination by an experienced examiner between 2016 and 2019. Patients recruited in Center 1 (Rome) were included as the training set and patients enrolled in Center 2 (Milan) formed the external validation set. Radiomics analysis (extraction of a high number of quantitative features from medical images) was applied to the ultrasound images. Clinical (including preoperative biopsy), ultrasound and radiomics features that were statistically significantly different in the high-risk group vs the other groups and in the low-risk group vs the other groups on univariate analysis in the training set were considered for multivariate analysis and for developing ultrasound-based machine-learning risk-prediction models. For discriminating between the high-risk group and the other groups, a random forest model from the radiomics features (radiomics model), a binary logistic regression model from clinical and ultrasound features (clinical-ultrasound model) and another binary logistic regression model from clinical, ultrasound and previously selected radiomics features (mixed model) were created. Similar models were created for discriminating between the low-risk group and the other groups. The models developed in the training set were tested in the validation set. The performance of the models in discriminating between the high-risk group and the other groups, and between the low-risk group and the other risk groups for both validation and training sets was compared. RESULTS: The training set comprised 396 patients and the validation set 102 patients. In the validation set, for predicting high-risk endometrial cancer, the radiomics model had an area under the receiver-operating-characteristics curve (AUC) of 0.80, sensitivity of 58.7% and specificity of 85.7% (using the optimal risk cut-off of 0.41); the clinical-ultrasound model had an AUC of 0.90, sensitivity of 80.4% and specificity of 83.9% (using the optimal cut-off of 0.32); and the mixed model had an AUC of 0.88, sensitivity of 67.3% and specificity of 91.0% (using the optimal cut-off of 0.42). For the prediction of low-risk endometrial cancer, the radiomics model had an AUC of 0.71, sensitivity of 65.0% and specificity of 64.5% (using the optimal cut-off of 0.38); the clinical-ultrasound model had an AUC of 0.85, sensitivity of 70.0% and specificity of 80.6% (using the optimal cut-off of 0.46); and the mixed model had an AUC of 0.85, sensitivity of 87.5% and specificity of 72.5% (using the optimal cut-off of 0.36). CONCLUSIONS: Radiomics seems to have some ability to discriminate between low-risk endometrial cancer and other endometrial cancers and better ability to discriminate between high-risk endometrial cancer and other endometrial cancers. However, the addition of radiomics features to the clinical-ultrasound models did not result in any notable increase in performance. Other efficacy studies and further effectiveness studies are needed to validate the performance of the models. © 2021 International Society of Ultrasound in Obstetrics and Gynecology.

Using rADioMIcs and machine learning with ultrasonography for the differential diagnosis of myometRiAL tumors (the ADMIRAL pilot study). Radiomics and differential diagnosis of myometrial tumors.

OBJECTIVE: To develop and evaluate the performance of a radiomics and machine learning model applied to ultrasound (US) images in predicting the risk of malignancy of a uterine mesenchymal lesion. METHODS: Single-center retrospective evaluation of consecutive patients who underwent surgery for a malignant uterine mesenchymal lesion (sarcoma) and a control group of patients operated on for a benign uterine mesenchymal lesion (myoma). Radiomics was applied to US preoperative images according to the International Biomarker Standardization Initiative guidelines to create, validate and test a classification model for the differential diagnosis of myometrial tumors. The TRACE4 radiomic platform was used thus obtaining a full-automatic radiomic workflow. Definitive histology was considered as gold standard. Accuracy, sensitivity, specificity, AUC and standard deviation of the created classification model were defined. RESULTS: A total of 70 women with uterine mesenchymal lesions were recruited (20 with histological diagnosis of sarcoma and 50 myomas). Three hundred and nineteen radiomics IBSI-compliant features were extracted and 308 radiomics features were found stable. Different machine learning classifiers were created and the best classification system showed Accuracy 0.85 ± 0.01, Sensitivity 0.80 ± 0.01, Specificity 0.87 ± 0.01, AUC 0.86 ± 0.03. CONCLUSIONS: Radiomics applied to US images shows a great potential in differential diagnosis of mesenchymal tumors, thus representing an interesting decision support tool for the gynecologist oncologist in an area often characterized by uncertainty.