Full-count PET Recovery from Low-count Image Using a Dilated Convolutional Neural Network.

Positron Emission Tomography (PET) is an essential technique in many clinical applications that allows for quantitative imaging at the molecular level. This study aims to develop a denoising method using novel dilated convolutional neural network to recover full-count images from low-count images. We adopted similar hierarchal structure from the conventional uNet and incorporated dilated kernels in each convolution to allow the network to observe larger, and perhaps, more robust, features within the image. Our dNet were trained alongside a uNet for comparison. Our 2.5D model used a training set (N=30) and testing set (N=5) that were obtained from an ongoing 18F-FDG study. Low-count PET data (10% count) were generated through Poisson thinning from the full listmode file. Both low-count PET and full-count PET were reconstructed with the OSEM algorithm. Objective imaging metrics including mean absolute percent error (MAPE), peak signal-to-noise ratio (PSNR) and structural similarity index metric (SSIM) were used to analyze the denoising methods. Both the uNet and our proposed dNet were successfully trained to synthesize full-count PET images from low-count PET images. Compared to low-count PET, both the uNet and dNet methods significantly improved MAPE, PSNR and SSIM. Our dNet also systematically outperformed uNet on all three metrics and across all testing subjects. This study proposed a novel approach of using dilated convolutions for recovering full-count PET images from low-count PET images. Our dNet significantly outperformed the well-established uNet and demonstrates great potential for denoising low-count PET images.