CBAM-Enhanced Faster R-CNN for Esophageal Cancer Detection in Barium Meal Imaging

Abstract

Esophageal cancer is a prevalent malignant tumor that poses a serious threat to human health. China has one of the highest incidences of esophageal cancer worldwide, with a large number of new cases diagnosed annually. Currently, the diagnosis of esophageal cancer relies primarily on electronic gastroscopy, which enables observation and detection through fluorescent screen imaging. As the number of patients continues to rise, the workload and pressure on physicians have increased substantiall. At the molecular level, the occurrence and development of esophageal cancer, like other cancers, are associated with the activation of proto-oncogenes and the inhibition of apoptosis-related genes. To address the problems in barium meal imaging where the esophageal region is inconspicuous and the background region occupies a large proportion of the feature map extracted by the Faster R-CNN backbone network, this study proposes CBAM Faster R-CNN. The convolutional block attention module (CBAM) is integrated into the original Faster R-CNN model to enhance the saliency of esophageal region features in the feature map. The CBAM Faster R-CNN model is trained on a data-enhanced training set, and Recall, Precision, and AP values are used for evaluation and analysis.