Crafting Synthetic Dental Radiographs with Generative Adversarial Networks

Artificial Intelligence has the potential to revolutionize dental radiology practices. However, training AI models necessitates precise and dependable training data, which can be challenging and costly to acquire. This study aims to generate synthetic or artificially generated data using a generative artificial intelligence technique. Specifically, it investigates synthetic radiography's utility in enhancing the clinical value of panoramic dental radiographs generated by Generative Adversarial Networks (GANs), a generative AI technique. We apply Fréchet Inception Distance (FID) to assess the quality of radiographs generated, finding similarities with other high-resolution tasks. Evaluating radiographic clinical realism involves having radiologists differentiate among authentic and synthetic radiographs. Our results reveal that the synthetic dental radiographs are prone to be classified as authentic showing FID as 18.1%o, indicating progress yet there exists room for refinement to achieve true degree of realism.