© FUJIFILM Corporation

A Training Method to Build Characterization Models for Pulmonary Nodules in CT Images Using Radiology Reports


To reduce the cost of the training data construction, authors propose a pseudo-labeling approach for automatic characterization of pulmonary nodules, i.e., image labels automatically retrieved from radiology reports. In the experiments, the classifier trained using their pseudo-labeling approach achieves almost the same performance as the one trained on ground truth labels on images manually annotated by radiologists. In the future, this method will help to build a CAD system that performs more detailed characterization.

Figure 1. Previous approach [based on [1]] (Expert checks images and annotates labels)
Figure 2. Proposed approach [based on [1]] (pseudo-labels are automatically predicted based on radiology reports)
Figure 3. Performance comparison between proposed method versus manual labeling. [1]
Figure 4. Examples of the model output trained on proposed method. [1]
[1] Momoki, Yohei, et al. “Characterization of Pulmonary Nodules in Computed Tomography Images Based on Pseudo-Labeling Using Radiology Reports.” IEEE Transactions on Circuits and Systems for Video Technology 32.5 (2021): 2582-2591.


DOI: https://doi.org/10.1109/TCSVT.2021.3073021