Search

Objective: Discrimination of ovarian tumors is a necessary procedure for proper treatment. In this study, we developed a convolutional neural network model with a convolutional autoencoder (CNN-CAE) to classify ovarian tumors. Methods: A total of 1,613 ultrasound images of ovarian tumors with the known pathologic diagnosis were pre-processed and augmented for deep learning analysis. We designed the CNN-CAE model that removes the unnecessary information on ultrasound images (e.g. calipers and annotations) and classifies ovarian tumors into five classes. We assessed the CNN-CAE model performance using a validation set of 327 images with the following metrics: accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). Gradient-weighted class activation mapping (Grad-CAM) was applied to visualize and verify the CNN-CAE model results qualitatively. Results: In classifying normal versus ovarian tumors, the CNN-CAE model showed 96.0% accuracy, 96.5% sensitivity, and 0.9837 AUC with DenseNet161 CNN architecture. In determining malignancy of ovarian tumors, the CNN-CAE model showed 86.85% accuracy, 75.68% sensitivity, and 0.9225 AUC with DenseNet121 CNN architecture. The Grad-CAM results show that the CNN-CAE model finds valid texture and morphology features from the ultrasound images and classifies ovarian tumors from these features. Conclusion: CNN-CAE is a feasible diagnostic tool that is capable of robustly classifying ovarian tumors by eliminating marks on ultrasound images. CNN-CAE demonstrates a significant application value in clinical conditions. Citation Format: Youn Jin Choi, Yuyeon Jung. Ovarian tumor diagnosis using deep convolutional neural networks and denoising convolutional autoencoder [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1925.