Material decomposition of spectral CT images via attention-based global convolutional generative adversarial network

Spectral computed tomography(CT)based on photon counting detectors can resolve the energy of every single photon interacting with the sensor layer and be used to analyze material attenuation information under different energy ranges,which can be helpful for material decomposition studies.However,there is a considerable amount of inherent quantum noise in narrow energy bins,resulting in a low signal-to-noise ratio,which can consequently affect the material decomposition performance in the image domain.Deep learning technology is currently widely used in medical image segmentation,denoising,and recognition.In order to improve the results of material decomposition,we propose an attention-based global convolutional generative adversarial network(AGC-GAN)to decompose different materials for spectral CT.Specifically,our network is a global convolutional neural network based on an attention mechanism that is combined with a generative adversarial network.The global convolutional network based on the attention mechanism is used as the generator,and a patchGAN discriminant network is used as the discriminator.Meanwhile,a clinical spectral CT image dataset is used to verify the feasibility of our proposed approach.Extensive experimental results demonstrate that AGC-GAN achieves a better material decomposition performance than vanilla U-Net,fully convolutional network,and fully convolutional denseNet.Remarkably,the mean intersection over union,structural similarity,mean precision,PAcc,and mean F1-score of our method reach up to 87.31%,94.83%,93.22%,97.39%,and 93.05%,respectively.

A study on noise reduction for dual-energy CT material decomposition with autoencoder

Purpose A major challenge for the material decomposition task of the dual-energy computed tomography(DECT)is the algorithm often suffers from heavy noise in the results.The purpose of this study is to propose a scheme to increase the noise performance of material decomposition.Methods The scheme we propose in this paper is to apply an autoencoder-based denoising procedure to the photon-counting DECT images before they are fed into the material decomposition algorithm.We implement the autoencoder(AE)by stacking a series of convolutional and deconvolutional layers.The decomposition technique adopted in our work is an iterative method using least squares estimation with the Huber loss function.The noises of the input and the output of material decomposition are analyzed with both simulated data and real data.Phantom and chicken wing experiments are conducted with a photoncounting-based spectral CT scanner to evaluate the proposed material decomposition scheme.Results The noise analysis of the input and the output of material decomposition demonstrates a positive correlation between them.Comparative experiment indicates a noise reduction in the output density maps for 26.07%to 35.65%after the autoencoder pre-processing is applied.The resultant contrast-to-noise ratio is largely increased,correspondingly.Conclusions By utilizing the additional autoencoder denoising step,the material decomposition algorithm achieves an improvement in the noise performance of the resultant density maps.

Precipitation of flaky moolooite and its thermal decomposition

Moolooite particles with flaky morphology were synthesized by mixing dilute solutions of copper nitrate and sodium oxalate in the presence of citric acid. Solution p H value, citric acid concentration, and stirring were found to have large effect on the shape of the precipitated particles. Under the stirring, the radial area of flaky moolooite particles was enlarged and extended to become a thinner and larger flake. This is ascribed to growth promotion caused by the selective absorption of citric ligands onto a particular crystalline surface of the moolooite particles. Flaky shape of the moolooite particles tended to become spherical and disappeared completely when decomposed under an Ar atmosphere, leading to the formation of large porous aggregated particles composed of many tiny nanosized copper crystals.