International audienceLow-dose CT is an effective solution to alleviate radiation risk to patients, it also introduces additional noise and streak artifacts. In order to maintain a high image quality for low-dose scanned CT data, we propose a post-processing method based on deep learning and using 2-D and 3-D residual convolutional networks. Experimental results and comparisons with other competing methods show that the proposed approach can effectively reduce the low-dose noise and artifacts while preserving tissue details. It is also pointed out that the 3-D model can achieve better performance in both edge-preservation and noise-artifact suppression. Factors that may influence the model performance, such as model width, depth, and dropout, are also examined

Sparse-view Reconstruction can be used to provide accelerated low dose CT imaging with both accelerated scan and reduced projection/back-projection calculation. Despite the rapid developments, image noise and artifacts still remain a major issue in the low dose protocol. In this paper, a deep learning based method named Improved GoogLeNet is proposed to remove streak artifacts due to projection missing in sparse-view CT reconstruction. Residual learning is used in GoogLeNet to study the artifacts of sparse-view CT reconstruction, and then subtracts the artifacts obtained by learning from the sparse reconstructed images, finally recovers a clear correction image. The intensity of reconstruction using the proposed method is very close to the full-view projective reconstructed image. The results indicate that the proposed method is practical and effective for reducing the artifacts and preserving the quality of the reconstructed image.

No abstract

The reactions of CCN(X̃2Π i ) radicals with normal alkanes have been studied at about 10 Torr total pressure and room temperature using the pulse laser photolysis/laser-induced fluorescence technique. CCN (X̃2Π i ) radicals are generated by the fourth harmonic of a Nd:YAG laser photolysis of Cl3CCN at 266 nm. The relative concentration of CCN (X̃2Π i ) radicals was monitored in the (0−0) band of the CCN (Ã2Δ ← X̃2Π i ) transition at 470.9 nm by laser induced fluorescence (LIF). From the analysis of the relative concentration−time behavior of CCN (X̃2Π i ) under pseudo-first-order conditions, the rate constants for the reaction of CCN (X̃2Π i ) with a series of normal alkanes (C1−C8) were determined for the first time. The new data establish that the gas-phase reactivity of small alkanes (C1−C8) towards the CCN radicals follows the linear free energy relationship typical of hydrogen abstraction. A comparison with the corresponding reactions of CN and OH radicals with a series of normal alkanes, leads us to suggest that the reaction of CCN (X̃2Π i ) with small normal alkanes proceeds via the mechanism of hydrogen abstraction. The plausibility of the suggested reaction mechanism is strengthened by bond dissociation energy (BDE) correlations and linear free energy correlations.

Accurate and timely landslide susceptibility mapping (LSM) is essential to effectively reduce the risk of landslide. In recent years, deep learning has been successfully applied to landslide susceptibility assessment due to the strong ability of fitting. However, in actual applications, the number of labeled samples is usually not sufficient for the training component. In this paper, a deep neural network model based on semi-supervised learning (SSL-DNN) for landslide susceptibility is proposed, which makes full use of a large number of spatial information (unlabeled data) with limited labeled data in the region to train the mode. Taking Jiaohe County in Jilin Province, China as an example, the landslide inventory from 2000 to 2017 was collected and 12 metrological, geographical, and human explanatory factors were compiled. Meanwhile, supervised models such as deep neural network (DNN), support vector machine (SVM), and logistic regression (LR) were implemented for comparison. Then, the landslide susceptibility was plotted and a series of evaluation tools such as class accuracy, predictive rate curves (AUC), and information gain ratio (IGR) were calculated to compare the prediction of models and factors. Experimental results indicate that the proposed SSL-DNN model (AUC = 0.898) outperformed all the comparison models. Therefore, semi-supervised deep learning could be considered as a potential approach for LSM.