Fig 2. CNN prediction of a representative tooth (4^th pre-molar).

(a) Datasets of deep learning model training, validation, and test. (b) The ground truth image reconstructed from the original 128-channel dense array. (c) Undersampled 64-channel imaging (left) versus its artifact-free output (right) predicted by the CNN model. (d) Undersampled 16-channel imaging (left) versus its artifact-free output (right) predicted by the CNN model. (e) One-dimensional line spread function profiles across the reconstructed tooth surface extracted from 128-channel ground truth, 64-channel predicted, and 16-channel predicted image. FWHM: full width at half maximum. Ground truth, input, and predicted images of the representative tooth at posterior position (f) and anterior position (g).