Radiomic Model Enhances Myometrial Invasion Prediction in Endometrial Cancer

A radiomic-based model has been proposed to predict the depth of myometrial invasion in endometrial cancer based on ultrasound images. Endometrial carcinoma is a prevalent tumor among women in Europe, with myometrial invasion being a crucial prognostic factor. The accuracy of preoperative assessment using radiological imaging is subjective and operator-dependent. The radiomic model developed in this study aims to standardize and enhance the accuracy of predicting myometrial invasion.

The study collected clinical data and ultrasound indicators from 77 patients with endometrial carcinoma. Pre-processing of ultrasound images was done, followed by using a pre-trained Inception-V3 convolutional neural network for feature extraction. Two models were trained: a binary classification model and a multiclass model, with a double step feature selection process. The models were validated through cross-validation rounds.

The binary model achieved an optimal classification performance with an AUC value of 90.76%. When the operator was assisted by the radiomic-based decision-making system, there was a significant increase in precision. The multiclass model also performed well, with an overall accuracy of 84.93% and good balance across per-class metrics. The operator’s performance was notably enhanced with the support of the radiomic model.

This study’s findings suggest that the radiomic-based models have the potential to improve the accuracy and standardization of assessing myometrial invasion in endometrial cancer using ultrasound images. The models can serve as valuable tools for clinicians, providing more precise estimates of myometrial infiltration during the diagnosis phase. Additionally, the models offer a low-cost and easily accessible examination method that causes no discomfort to the patient, making them highly beneficial in clinical practice.

While the study shows promising results, further validation on larger datasets and optimization of the models are needed to enhance their performance and applicability in real-world clinical settings. The proposed models represent a significant advancement in predicting myometrial invasion in endometrial cancer, offering a reliable and standardized approach to improve patient management and treatment decisions.