Cephalogram synthesis and landmark detection in dental cone-beam CT systems

Due to the lack of a standardized 3D cephalometric analysis methodology, 2D cephalograms synthesized from cone-beam computed tomography (CBCT) volumes are widely used for in dental CBCT systems. However, compared with conventional X-ray film based cephalograms, such synthetic image contrast and resolution, which impairs landmark identification. In addition, increased radiation dose applied acquire scan reconstruction causes potential health risks. this work, we propose sigmoid-based intensity transform that uses nonlinear optical property films increase volumes. To improve super resolution deep learning techniques investigated. For low purpose, pixel-to-pixel generative adversarial network (pix2pixGAN) is proposed cephalogram synthesis directly two projections. detection an efficient automatic method using combination LeNet-5 ResNet50 proposed. Our experiments demonstrate efficacy pix2pixGAN synthesis, achieving average peak signal-to-noise ratio (PSNR) value 33.8 reference Pix2pixGAN also achieves best performance PSNR 32.5 without introduction checkerboard or jagging artifacts. 86.7% successful rate 2 mm clinical acceptable range on ISBI Test1 data, comparable state-of-the-art methods. The trained can be 93.0% 80.7% 4 precision projections, respectively.