Abstract
The acquisition of improved quality via the administration of low radioactivity is an important challenge in the field of nuclear medicine. Thus, we evaluated the image quality generated under low radioactivity using deep-learning algorithms. In addition, internal dosimetry for various organs was measured based on Monte Carlo simulation. The dataset comprised 500 slices of thyroid nuclear medicine images acquired at 370 MBq as high radioactivity and 18.5 MBq as low radioactivity. The residual neural network (ResNet) with 16 residual blocks and a U-Net model were used with various hyperparameters for learning. In addition, we measured the internal dosimetry in peripheral organs based on a Monte Carlo simulation. The peak signal-to-noise ratio (PSNR) results were 18.24 for the ResNet model and 18.85 for the U-Net model on average. The structural similarity index measurement (SSIM) results were also 0.995 and 0.994 for the ResNet and U-Net models, respectively. The average absorbed dose for radionuclides with 37 MBq was 3.98 times higher than that of the radionuclide with 18.5 MBq. In conclusion, it is possible to make a diagnosis based on the administration of low radioactivity when scanning thyroid nuclear medicine images.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call