Akademik Veri Yönetim Sistemi
Signal, Image and Video Processing, cilt.20, sa.3, 2026 (SCI-Expanded, Scopus)
Dental panoramic imaging, which allows for the comparision all teeth, is a frequently used imaging method by dentists. This technique enables detailed intraoral assessment with a single X-ray, aiding in both diagnosis and treatment. In this study, we used two publicly available datasets, TUFTS and OdontoAI, which contain panoramic X-ray images. These datasets include a total of 3000 panoramic X-ray images. Within the scope of the study, new images were produced from existing panoramic images in an anonymous and ethically compliant manner. In the study, the performance of the pix2pixHD, CLADE and SPADE models on the generated images was investigated. The SPADE model generates images by directly incorporating conditions into the generator layers, effectively utilizing the latent space. This approach enables more efficient use of computational resources during the image synthesis process. To improve the performance of the SPADE model, various modifications were made to the model blocks, leading to the development of three new models: SPADE2XGRNX, SPADE2XGRNX Symmetry and SPADE2XGRNX SymmetryInBlocks. The effectiveness of the proposed models was evaluated using the parameters RMSE of 0.0074, SSIM of 0.9541, PSNR of 42.69, FID of 49.06, and KID of 0.015 on both the OdontoAI and TUFTS datasets. The results demonstrate the effectiveness of the proposed models. Additionally, a real-fake scoring was obtained for the synthetic images by 24 dentists. It has been observed that the dentists found some of the images generated by the newly proposed models to be realistic.