Aiming at the problem that detection rate of lung nodules detection scheme based on rabat is low and has a lot of false positives

propose a new nodules detection scheme based on convolutional neural network（ CNN）.In the scheme

enhance chest radiograph

and then pick positive and negative samples to train the CNN. Process the enhanced image using sliding windows method with the pre-trained network to get the region of interest（ ROI）

and exclude the false positives by using the size of the ROI at last. The proposed scheme omits the procedure of segmentation of lung field in traditional schemes. And this can avoid losing nodules which are excluded by the segmentation procedure. The JSRT database is used to evaluate the system. The scheme achieves a sensitivity of86 % for all nodules and a detection rate of 84 % with 5. 0 FPs per radiograph for very subtle and extremely subtle nodules which outperform the current reported methods.