In order to solve the problem of low detection sensitivity and high false positives of traditional lung nodule detection models, a multi-scale three dimensional (3D) convolutional neural network (CNN) based lung nodule detection method was proposed. First, in order to improve the operating speed and network flexibility, a single-stage mode was adopted, and there was no false positive reduction stage. Secondly, based on the above model, we built a new network structure, innovatively used the 3D UNet++-like architecture as the backbone of the region proposal network (RPN), and adopted the flexible nesting mode of residual blocks. Finally, the three input sizes were input into the 3D neural network, and their classification results were merged. Experiments showed that our model had an average sensitivity of 87.3% in false positive screening based on the LUNA16 datasets, which was an increase of 7.8% compared with the UNet++ network. The total sensitivity was as high as 96.2%. It can be seen that our model can significantly improve detection sensitivity and reduce false positives, which can provide a theoretical reference for clinical applications.