Deep Learning Prediction of Ovarian Malignancy at US Compared with O-RADS and Expert Assessment

• Hui Chen*, 

• Bo-Wen Yang*, 

• Le Qian, 

• Yi-Shuang Meng, 

• Xiang-Hui Bai, 

• Xiao-Wei Hong, 

• Xin He, 

• Mei-Jiao Jiang, 

• Fei Yuan, 

• Qin-Wen Du, 

• Wei-Wei Feng 

* H.C. and B.W.Y. contributed equally to this work.

Published Online:Apr 12 2022

A deep learning model that used feature fusion to classify benign and malignant ovarian tumors on gray scale and color Doppler US images achieved a similar performance to clinical expert assessment.

Background

Deep learning (DL) algorithms could improve the classification of ovarian tumors assessed with multimodal US.

Purpose

To develop DL algorithms for the automated classification of benign versus malignant ovarian tumors assessed with US and to compare algorithm performance to Ovarian-Adnexal Reporting and Data System (O-RADS) and subjective expert assessment for malignancy.

Materials and Methods

This retrospective study included consecutive women with ovarian tumors undergoing gray scale and color Doppler US from January 2019 to November 2019. Histopathologic analysis was the reference standard. The data set was divided into training (70%), validation (10%), and test (20%) sets. Algorithms modified from residual network (ResNet) with two fusion strategies (feature fusion [hereafter, DLfeature] or decision fusion [hereafter, DLdecision]) were developed. DL prediction of malignancy was compared with O-RADS risk categorization and expert assessment by area under the receiver operating characteristic curve (AUC) analysis in the test set.

Results

A total of 422 women (mean age, 46.4 years ± 14.8 [SD]) with 304 benign and 118 malignant tumors were included; there were 337 women in the training and validation data set and 85 women in the test data set. DLfeature had an AUC of 0.93 (95% CI: 0.85, 0.97) for classifying malignant from benign ovarian tumors, comparable with O-RADS (AUC, 0.92; 95% CI: 0.85, 0.97; P = .88) and expert assessment (AUC, 0.97; 95% CI: 0.91, 0.99; P = .07), and similar to DLdecision (AUC, 0.90; 95% CI: 0.82, 0.96; P = .29). DLdecision, DLfeature, O-RADS, and expert assessment achieved sensitivities of 92%, 92%, 92%, and 96%, respectively, and specificities of 80%, 85%, 89%, and 87%, respectively, for malignancy.

Conclusion

Deep learning algorithms developed by using multimodal US images may distinguish malignant from benign ovarian tumors with diagnostic performance comparable to expert subjective and Ovarian-Adnexal Reporting and Data System assessment.

© RSNA, 2022