Evaluation of a Pulmonary Nodule Detection Performance on Chest Radiograph by Mask R-CNN

Because of the traditional computer aided detection (CADe) scheme is difficult to distinguish some real nodules and bone structures based on the features extracted from experience, which gives high false positives (FPs) in the detection results, we introduce mask region-baseel convolutional neural network (Mask R-CNN) model for nodule detection in Chest X-Ray radiograph (CXR). We apply the limited contrast adaptive histogram equalization (CLAHE) and a multi-segment active shape model (M-ASM) to enhancing the nodule in CXR. The model selects ResNet-101 and feature pyramid network (FPN) as the backbone to extract image features. According to the test of the Japanese Society of Radiological Technology (JSRT) public database, the proposed scheme achieves a sensitivity of 93.57% with an average of 5.0 false positives (FPs) and 78% with an average of 2.0 FPs. Compared with other methods, most of the pulmonary nodules outside the lung area are detected in our scheme (10/14).