An Efficient Approach for Gastric Polyps Detection Based on Improved SSD

Gastroscopy is the main method of detecting gastric polyps. Specifically, the doctor observes the gastroscope image with the naked eye to determine the presence and location of polyps. The detection rate of polyps is affected by the doctor’s experience, fatigue, and other factors. There is a certain degree of missed polyps. Therefore, it is necessary to detect polyps by auxiliary means. In this paper, we report a convolutional neural network for gastric polyp detection, which is based on the Single Shot MultiBox Detector (SSD). In order to improve the performance of polyp detection, an excellent feature extractor is essential. We adopt EfficientNet instead of VGG16 as the backbone network of the SSD. In addition, the previous loss function is improved, and the PReLU activation function is introduced into the SSD to achieve better detection performance. We conduct extensive experiments on the gastric polyp dataset. The final experimental results show that the improved SSD network for gastric polyp detection can achieve a precision of 90.3% and a recall rate of 85.7%, which is better than the original SSD network.