Lung nodule detection method based on dual path fusion feature pyramid

The detection of pulmonary nodules has important clinical significance. Aiming at the characteristics of pulmonary nodule types, size diversity and location randomness, this paper proposes a pulmonary nodule detection method based on dual-path fusion feature pyramid. First, a bottom-up fusion path is added on the basis of the feature pyramid structure, and the shallow feature flow is realized through very few convolutional layers, and the target positioning ability is improved. Then the context attention mechanism is introduced in the feature fusion stage to improve the feature expression of the region of interest. Finally, a residual structure combined with depth separable convolution is proposed to reduce the time cost and make the model more lightweight. Experiments show that the proposed method can effectively extract the multi-scale features of nodules, with an average sensitivity of 96.18% and a CPM value of 86.34%, achieving a low number of false positives while ensuring a high recall rate of nodules. The detection results of this method are improved compared with previous methods, and it has certain clinical value.