Expressive feature representation pyramid network for pulmonary nodule detection

Lung cancer has the highest fatality rate among all types of cancers. The detection of pulmonary nodules serves as the primary means for early diagnosis, utilizing deep learning models for pulmonary nodule detection can improve the accuracy and efficiency of detection. However, existing feature extraction networks fail to capture precise details and shape characteristics of pulmonary nodules, and they also lack sufficient multi-scale fusion. Therefore, we propose the expressive feature representation pyramid (EFRP) for pulmonary nodule detection. The Context Enhancement Connection module generates more discriminative features by performing three scales of context feature extraction through different paths and utilizes rich local information and global contextual information to enhance feature representation. The Adaptive Feature Enhancement module dynamically adjusts the receptive field size and generates multi-scale feature layers with enhanced features. The Channel Attention Feature Refinement module enhances local interactions between different channels to alleviate the mixed effects caused by the fusion process, thereby increasing the robustness of the model. Through extensive experiments on three different publicly available pulmonary nodule datasets, the results demonstrate EFRP not only ensure precision but also reduce the occurrence of missed detections, effectively enhancing the overall detection performance of pulmonary nodules.