Novel SSD-Based Models for Detection of Multidimensional Pulmonary Nodules in CT Images

Lung cancer, as one of the diseases with the highest incidence and mortality rates in the world, has always been a big challenge that humans are constantly striving to address. Pulmonary nodules, as one of the early manifestations of lung cancer, constitute a breakthrough in early detection of this disease and its prompt treatment. Currently, significant progress has been made in object detection models based on deep learning in the context of increasingly scarce medical resources. To make pulmonary nodule detection more efficient and accurate, this paper proposes two novel models, based on the Single Shot MultiBox Detector (SSD) model. The proposed models adapt to the detection of small pulmonary nodules by changing the size of the feature map of the input prediction module. In addition, the prediction frame is modified to make greater use of the shallow feature layer. A group of up-sampling modules is added by increasing asymmetric convolution and a feature pyramid network (FPN) mechanism is used for improving the detection performance. The experimental results, obtained on the LUNA16 dataset, demonstrate that the proposed models achieve great performance improvement compared to the baseline (SSD).