Pulmonary nodule detection on chest radiographs using balanced convolutional neural network and classic candidate detection

作者：

Highlights：

• A combination of the classic nodule candidate detection method and the CNN method for false positive reduction is presented to achieve high performance for lung nodule detection in CXR.

• The nodule enhancement ROI image, corresponding segmentation result, and original ROI image were encoded into an RGB color image instead of the duplicated original ROI image as the input of the CNN for false positive reduction.

• All sub modules are designed for processing the original size of CXR to avoid missing small nodules when down sampling the CXR.

摘要

•A combination of the classic nodule candidate detection method and the CNN method for false positive reduction is presented to achieve high performance for lung nodule detection in CXR.•The nodule enhancement ROI image, corresponding segmentation result, and original ROI image were encoded into an RGB color image instead of the duplicated original ROI image as the input of the CNN for false positive reduction.•All sub modules are designed for processing the original size of CXR to avoid missing small nodules when down sampling the CXR.

论文关键词：Computer-aided detection (CADe),Transfer learning,Chest radiograph (CXR),Pulmonary nodule

论文评审过程：Received 14 August 2019, Revised 5 April 2020, Accepted 12 May 2020, Available online 22 May 2020, Version of Record 23 June 2020.

论文官网地址：https://doi.org/10.1016/j.artmed.2020.101881