Skip to main content
. 2023 Oct 30;18(10):e0293468. doi: 10.1371/journal.pone.0293468

Fig 1. The overview of deep learning-assisted sparse array periodontal imaging.

Fig 1

(a) Schematics of imaging periodontal anatomy using an ultrasound transducer. The ultrasound transducer includes an array of elements of a specific pitch (i.e., spacing). Periodontal structural anatomy includes the tooth, gingiva, alveolar bone crest (ABC), cementoenamel junction (CEJ), and gingival margin (GM). (b) Periodontal imaging reconstructed from 128-, 64- and 16-channel ultrasound transducer. By regularly turning off every other channel of the original 128-channel transducer, the pitch of 64-channel is 140 μm. Similarly, by regularly turning off seven out of eight channels of the original 128-channel transducer, the pitch of 16-channel is 560 μm. (c) The CNN-based RUNet architecture that predicts high-quality images from the sparse array imaging input. Conv2D: convolutional layer. BN: batch norms. ReLU: rectified linear units.