Deep Learning Driven Noise Reduction for Reduced Flux Computed Tomography

Deep neural networks have received considerable attention in clinical imaging, particularly with respect to the reduction of radiation risk. Lowering dose by reducing photon flux inevitably results degradation scanned image quality. Thus, researchers sought exploit deep convolutional (DCNNs) map low-quality, low-dose images higher-dose, higher-quality images, thereby minimizing associated hazard. Conversely, computed tomography (CT) measurements geomaterials are not limited dose. In contrast human body, however, may be comprised high-density constituents causing increased attenuation X-rays. Consequently, higher-dose required obtain an acceptable scan The problem prolonged acquisition times is severe for micro-CT based scanning technologies. Depending on sample size and exposure time settings, a single require several hours complete. This particular concern if phenomena exponential temperature dependency elucidated. A process happen too fast adequately captured CT scanning. To address aforementioned issues, we apply DCNNs improve quality rock reduce more than 60%, simultaneously. We highlight current derived datasets transfer learning DCNN without increasing training time. approach applicable any technology. Furthermore, performance trained different loss functions such as mean squared error structural similarity index.