Deep learning, the mainstream of artificial intelligence (AI), has made progresses in computer vision, exacting information of multi-scale features from images. Since 2016, deep learning methods are being actively developed for tomography, reconstructing images of internal structures from their integrative features such as line integrals. There are both excitements and challenges in the Wild West of AI at large, and AI-based imaging in particular, involving accuracy, robustness, generalizability, interpretability, among others. Based on the author’s plenary speech at SPIE Optics + Photonics, August 2, 2021, here we provide a background where x-ray imaging meets deep learning, describe representative results on low-dose CT, sparse-data CT, and deep radiomics, and discuss opportunities to combine datadriven and model-based methods for x-ray CT, other imaging modalities, and their combinations so that imaging service can be significantly improved for precision medicine.
|