Characterization of Pulmonary Nodules in Computed Tomography Images Based on Pseudo-Labeling Using Radiology Reports

A computer-aided diagnosis (CAD) system that characterizes nodules in medical images can help radiologists determine its malignancy. Preparing large volumes of labeled data for CAD systems, however, requires advanced knowledge. This makes it extremely difficult to develop such despite their growing demand. In this paper, we propose a new training method build an image classifier characterization utilizing pseudo-labels, i.e., labels automatically retrieved from radiology reports. report is type record which present summary lesion characteristics and diagnosis. Labeling reports much easier than labeling images, be done without high expertise. Using several thousand reports, constructed hierarchical attention network-based text assign pseudo-labels the pulmonary with accuracy (macro F1-score 0.941). Experimental results show trained achieve almost same performance as one annotated by radiologists: AUC 0.848 model on 3,000 computed tomography (CT) 0.847 manual 800 CT images.