Fig. 4: The proposed workflow of the multi-view-based image classification for foreign body aspiration detection.
From: Automated detection of radiolucent foreign body aspiration on chest CT using deep learning
In brief, the CT images undergo preprocessing and airway tree extraction to generate 3D airway models. Multi-snapshots of these models are taken from different angles. These snapshots are then processed using a convolutional neural network (CNN) architecture, which includes convolution, max pooling, and fully connected layers. Finally, the processed images are classified into two categories: FBA (foreign body aspiration) and NFBA (non-foreign body aspiration). De-identified CT images, identical to those presented in Fig. 2, are shown with the informed consent requirement waived by the corresponding Ethics Review Committee.
