A Pulmonary Nodule Detection Algorithm Based on Low Dose CT Images

In order to accurately detect the location of pulmonary nodules in hundreds of chest CT images in routine reading environment, this paper proposes an improved algorithm based on Faster R-CNN. Firstly, we concatenate multi-level feature maps in VGG16 model to fuse the shallow and deep features of the shared convolution layer, which recovers the more fine-grained features. Then, we design a new "Pyramid RPN" structure with three parallel convolution kernels of different sizes to generate more accurate candidate regions. Finally, the region of interest (ROI) pooling layer is optimized by removing quantization operations and using bilinear interpolation to compute the exact value to reduce regression deviation. The experimental results show that the sensitivity and the false positive rate of each scan have a better performance improvement. The proposed method can more accurately detect small pulmonary nodules and has certain clinical significance for early screening of lung cancer.