Feature Fusion-based Automatic Grading of Color Doppler Flow within Ovarian Masses in Ultrasound Images.

Ovarian cancer is the gynecological malignant tumor with the highest mortality. The early diagnosis still faces challenges. With the rapid development of computer-aided diagnosis technology, the demand for application in the medical industry has greatly increased. According to the Ovarian-adnexal Reporting and Data System (O-RADS), we propose an automatic grading approach for color Doppler flow within ovarian masses. After separating the color Doppler flow from the original image, the features such as histogram of oriented gradient (HOG), gray-level co-occurrence matrix (GLCM), and their fused features are extracted. The stratified 10-fold cross-validation is adopted to train the support vector machine (SVM) and k-nearest neighbor (KNN) classifiers. Experiments on a dataset with 570 ovarian color Doppler ultrasound images generate an average accuracy of 92.64%, a recall of 90.30%, a specificity of 97.75%, and a macro_F1 of 90.07%. The results show that the fused feature of GLCM and HOG can better grade the blood flow compared with other features. KNN has superior performance over SVM for this task. The color score of color Doppler flow within ovarian masses combined with the B-mode ultrasound images can ensure the integrity of key information while avoiding redundancy. The study lays the foundation for the development of a subsequent O-RADS grading system for ovarian masses.