Title: AT-Cascade R-CNN: a novel attention-based cascade R-CNN model for ovarian cancer lesion identification

Authors: Shudong Wang; Lisheng Wang; Lei Wang; Zhiyong Yu; Xiumin Zhao; Xun Wang

Addresses: College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580, Shandong, China ' College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580, Shandong, China ' Department of Gynaecology, The Affiliated Hospital of Qingdao University, Qingdao, 266580, Shandong, China ' College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580, Shandong, China ' Department of Neurology, Shandong Provincial the Third Hospital, Shandong University, Jinan, 250100, Shandong, China ' College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580, Shandong, China

Abstract: Ovarian cancer is one of the most common cancers in female genital organs. Since ovarian cancer is associated with plenty of pathological features, most of the computational diagnostic methods can classify the type of ovarian cancer, but cannot accurately detect the locations. In this work, we propose a novel deep learning model architecture, named AT-Cascade R-CNN. In our model, attention mechanism is firstly involved into Cascade R-CNN. We use attention-based activation function in the backbone network to achieve nonlinear activation. Data experiments are conducted on dataset collected from the affiliated hospital of Qingdao University. After training, our model performs better than baseline Cascade R-CNN and Cascade R-CNN+CBAM in detecting the lesions. The MAP of our model reaches 93.2%, which is better than some state-of-the-art deep learning models. It realises accurate detection of ovarian cancer lesions for the first time, and accurately classifies the lesions.

Keywords: ovarian cancer; cascade R-CNN; attention mechanism.

DOI: 10.1504/IJAIS.2021.117913

International Journal of Adaptive and Innovative Systems, 2021 Vol.3 No.1, pp.74 - 86

Received: 12 Apr 2021
Accepted: 23 Apr 2021
Published online: 04 Oct 2021 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article