Lung adenocarcinoma diagnosis in one stage

Abstract Early detection of lung cancer is the most promising path to increase the chance of survival for patients. Accurate lung nodule detection in computed tomography (CT) images is a crucial step in diagnosing lung cancer. CNN based CADe systems diagnose lung nodules in two stages by using one network for detection and another following network for classification or false positive reduction. Besides costly construction, two-stage systems can only find a single-scale location including many surrounding contexts for each nodule, which leads to a limited accuracy when classifying small nodules. In this paper, we construct a one-stage framework relying on feature pyramid network (FPN) for the diagnosis of lung adenocarcinoma (LA). The proposed network has two advantages, (i) it can localize and classify LAs simultaneously. (ii) it generates feature maps with high resolution, from which robust classification is reached since tight coverage takes less contextual information. The performance of the proposed one-stage diagnostic network is verified on a lung adenocarcinoma dataset. Experimental results show that it achieves higher sensitives than two-stage frameworks.