Deep learning based coherence holography reconstruction of 3D objects.

We propose a fast, new, position-independent reconstruction method, to our knowledge, for coherence holography using deep learning (DL) method, dubbed fast deep coherence holography (FDCH). Our trained network was developed to reconstruct the complex object field at the location corresponding to the position of the camera, which opens the ability to apply angular spectrum propagation to reconstruct the 3D object. Phase-shift coherence holography is a traditional analytical technique that also reconstructs the complex object, but it is subject to noise such as experimental noise, camera noise. In a previous work, deep coherence holography (DCH) was developed to predict a non-diffracted field from a noisy image, but it requires the camera to be inserted at the right position of the non-diffracted field, which increases the complexity and the cost of the experimental setup. In addition, while the analytical method one requires Nsin (Nsin>3) captured interferograms to reconstruct the complex field, and thenumber of the recorded images of the DCH is proportional to the number of sub-objects contained in the object, our new proposed DL method only needs two images at a random camera position. Hence, the total reconstruction time is reduced. Not only does the FDCH reconstruct the 3D object faster at any random position, but it also yields a better result than the above methods in terms of accuracy.

The subband coding method holds a preeminent position for image compression. The discrete version of the wavelet transform is closely related to dyadic subband filter banks which have been used in image processing. Subband coding and vector quantization have been shown to be effective methods for coding images at low bit rates. We compute the mean squared reconstruction error (MSE) which depends on N, the number of the entries in each codebook, and k the length of each codeword (that is, the average bit rate) and on the subband filter bank coefficients. We form this MSE measure in terms of the equivalent quantization model and find the optimum FIR filter coefficients for each equivalent channel in the subband structure for a given bit rate, given filter length, and given input signal correlation model. Specific design examples are shown for 4-tap filter paraunitary filter bank structure. Theoretical results are confirmed by extensive Monte Carlo simulation.

As training a high-performance deep neural network (DNN) model requires a large amount of data, powerful computing resources and expert knowledge, protecting well-trained DNN models from intellectual property (IP) infringement has raised serious concerns in recent years. Most existing methods using DNN watermarks to verify the ownership of the models after IP infringement occurs, which is reactive in the sense that they cannot prevent unauthorized users from using the model in the first place. Different from these methods, in this article, we propose an active authorization control and user’s fingerprint tracking method for the IP protection of DNN models by utilizing sample-specific backdoor attack. The proposed method inversely and multiplely exploits sample-specific trigger as the key to implement authorization control for DNN model, in which the generated triggers are imperceptible and sample-specific for clean images. Specifically, a U-Net model is used to generate backdoor instances. Then, the target model is trained on the clean images and backdoor instances, which are inversely labeled as wrong classes and correct classes, respectively. Only authorized users can use the target model normally by pre-processing the clean images through the U-Net model. Moreover, the images processed by the U-Net model will contain unique fingerprint that can be extracted to verify and track the corresponding user’s identity. This article is the first work that utilizes the sample-specific backdoor attack to implement active authorization control and user’s fingerprint management for DNN model under black-box scenarios. Extensive experimental results on ImageNet dataset and YouTube Aligned Face dataset demonstrate that the proposed method is effective in protecting the DNN model from unauthorized usage. Specifically, the protected model has a low inference accuracy (1.00%) for unauthorized users, while maintaining a normal inference accuracy (97.67%) for authorized users. Besides, the proposed method can achieve 100% fingerprint tracking success rates on both the ImageNet and YouTube Aligned Face datasets. Moreover, it is demonstrated that the proposed method is robust against fine-tuning attack, pruning attack, pruning attack with retraining, reverse-engineering attack, adaptive attack, and JPEG compression attack. The code is available at https://github.com/nuaaaisec/SSAT .

We discuss a method of signal analysis — empirical mode decomposition, and also its modification — complementary ensemble empirical mode decomposition. Both methods are used to research the reconstruction of a speech signal by the means of intrinsic mode functions that were received during the decomposition. Researches were performed using two English databases of speech signals which contain speech sentences. Sentences are pronounced by male and female speakers. We utilized next objective indexes for our investigations: relative power and root mean square error of reconstruction, quality index PESQ and intelligibility index SNRloss. Results of a computational experiment are given. These results demonstrate a high quality of the signal reconstruction at its full restoration, and also advantages of the complementary decomposition in comparison with the customary decomposition.

We are developing a decision support system for treatment response assessment of bladder cancer in CT urography (CTU). Accurate segmentation of bladder cancer is a critical and challenging task. We previously developed a bladder cancer segmentation method using a deep learning convolutional neural network and level set (DL-CNN+LS) approach. In this study we investigated the application of a U-Net based deep learning (U-Net) model for bladder cancer segmentation. Our new U-Net method did not require the second-stage level set refinement, greatly simplifying the overall segmentation pipeline. The proposed U-Net model utilized a user-defined box to direct the attention of the U-Net to the lesion region by masking out the structured background outside the box. We trained and evaluated the performance of the U-Net segmentations by using hand-drawn 3D contours from a radiologist as reference standard. The segmentation accuracy was evaluated by the average volume intersection ratio (AVI), average percent volume error (AVE), average absolute volume error (AAVE), average minimum distance (AMD), and the Jaccard index (JI). On the validation set, the cropped U-Net achieved values of AVI = 65.5±20.2%, AVE = 3.7±44.4%, AAVE = 30.9±30.8%, AMD = 2.9±1.5 mm, and JI = 51.4±16.8%. The previous DL-CNN+LS approach achieved values of AVI = 36.2±28.4%, AVE = 51.1±39.3%, AAVE = 52.0±38.0%, AMD = 4.1±1.7 mm, and JI = 32.9±27.2%. On the independent test set, the cropped U-Net achieved values of AVI = 66.0±22.8%, AVE = -3.7±39.4%, AAVE = 28.1±27.3%, AMD = 4.5±3.3 mm, and JI = 49.2±20.0%. The DL-CNN+LS achieved values of AVI = 31.2±24.3%, AVE = 59.9±30.9%, AAVE = 60.8±29.0%, AMD = 5.5±2.2 mm, and JI = 27.9±20.8%, respectively. The results demonstrated that the U-Net model could achieve a higher accuracy than the previous DL-CNN+LS model while reducing the complexity of the segmentation pipeline.

Single-Cell Map of Childhood Acute Myeloid Leukaemia Using Variational Auto-Encoders

Style transfer is an optimizing technique that aims to blend style of input image to content image. Deep neural networks have previously surpassed humans in tasks such as object identification and detection. Deep neural networks, on the contrary, had been lagging behind in generating higher quality creative products until lately. This article introduces deep-learning techniques, which are vital in accomplishing human characteristics and open up a new world of prospects. The system employs a pre-trained CNN so that the styles of the provided image is transferred to the content image to generate high quality stylized image. The designed systems effectiveness is evaluated based on Mean Square Error (MSE), Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index Metrics (SSIM), it is noticed that the designed method effectively maintains the structural and textural information of the cover image.

Brain Magnetic Resonance image diagnostics employs image processing, but aberrations such as Intensity Inhomogeneity (IIH) distort the image, making diagnosis difficult. Clinical diagnostic methods must address IIH discrepancies in brain MR scans, which occur often. Accurate brain MR image processing is difficult but required for clinical diagnosis. In this study, we introduced a more energy-efficient intensity inhomogeneity correction (IIC) method that makes use of the Modified Energy-based Generative Adversarial Network. This method uses reconstruction error in the discriminator architecture to save energy by altering the cost function. The generator's performance is also improved by this reconstruction error. As the reconstruction error decreases, the discriminator collects latent information from real images to enhance output. To prevent mode collapse, the model has a drawing away term (PT). The generator design is improved by using skip connections and information modules that collect features at various scales. The suggested method beats state-of-the-art methods in metrics such as Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), Multi-Scale Structural Similarity Index (MSSSIM), Mean Squared Error (MSE), and Root Mean Square Error (RMSE).

A small dataset commonly affects generalization, robustness, and overall performance of deep neural networks (DNNs) in medical imaging research. Since gathering large clinical databases is always difficult, we proposed an analytical method for producing a large realistic/diverse dataset. Clinical brain PET/CT/MR images including full-dose (FD), low-dose (LD) corresponding to only 5 % of events acquired in the FD scan, non-attenuated correction (NAC) and CT-based measured attenuation correction (MAC) PET images, CT images and T1 and T2 MR sequences of 35 patients were included. All images were registered to the Montreal Neurological Institute (MNI) template. Laplacian blending was used to make a natural presentation using information in the frequency domain of images from two separate patients, as well as the blending mask. This classical technique from the computer vision and image processing communities is still widely used and unlike modern DNNs, does not require the availability of training data. A modified ResNet DNN was implemented to evaluate four image-to-image translation tasks, including LD to FD, LD+MR to FD, NAC to MAC, and MRI to CT, with and without using the synthesized images. Quantitative analysis using established metrics, including the peak signal-to-noise ratio (PSNR), structural similarity index metric (SSIM), and joint histogram analysis was performed for quantitative evaluation. The quantitative comparison between the registered small dataset containing 35 patients and the large dataset containing 350 synthesized plus 35 real dataset demonstrated improvement of the RMSE and SSIM by 29% and 8% for LD to FD, 40% and 7% for LD+MRI to FD, 16% and 8% for NAC to MAC, and 24% and 11% for MRI to CT mapping task, respectively. The qualitative/quantitative analysis demonstrated that the proposed model improved the performance of all four DNN models through producing images of higher quality and lower quantitative bias and variance compared to reference images.

In this paper, we propose a convex optimization framework for simultaneous estimation of super-resolved depth map and images from a single moving camera. The pixel measurement error in 3D reconstruction is directly related to the resolution of the images at hand. In turn, even a small measurement error can cause significant errors in reconstructing 3D scene structure or camera pose. Therefore, enhancing image resolution can be an effective solution for securing the accuracy as well as the resolution of 3D reconstruction. In the proposed method, depth map estimation and image super-resolution are formulated in a single energy minimization framework with a convex function and solved efficiently by a first-order primal-dual algorithm. Explicit inter-frame pixel correspondences are not required for our super-resolution procedure, thus we can avoid a huge computation time and obtain improved depth map in the accuracy and resolution as well as high-resolution images with reasonable time. The superiority of our algorithm is demonstrated by presenting the improved depth map accuracy, image super-resolution results, and camera pose estimation.

Traditional deblurring algorithms such as blur kernel estimation and gradient prior methods, these methods can only deal with a single type of blurry image, and such methods are very time-consuming. Due to the excellent performance of deep learning neural networks in the field of image generation, this paper regards the traditional image deblurring problem as a special image transformation problem. First, a multi-scale network is used to restore the blurry image, the network can restore the image in an end-to-end manner. Then, the multi-scale network will output a restored image. Finally, the generative adversarial network (GAN) can make the restored image closer to the real clear image in the sense. In terms of loss function, perceptual loss in content loss is adopted to ensure the consistency of the generated image and clear image in content. The results of deblurring will be evaluated by testing the peak signal to noise ratio (PSNR) and structural similarity (SSIM) between the generated image and the clear image. The results show that the PSNR and SSIM of the algorithm in the GoPro dataset have reached 28.01 and 0.8538, respectively, and can recover some details lost due to shake blur.

ObjectiveWe aimed to develop a deep neural network for segmenting lung parenchyma with extensive pathological conditions on non-contrast chest computed tomography (CT) images.Materials and MethodsThin-section non-contrast chest CT images from 203 patients (115 males, 88 females; age range, 31–89 years) between January 2017 and May 2017 were included in the study, of which 150 cases had extensive lung parenchymal disease involving more than 40% of the parenchymal area. Parenchymal diseases included interstitial lung disease (ILD), emphysema, nontuberculous mycobacterial lung disease, tuberculous destroyed lung, pneumonia, lung cancer, and other diseases. Five experienced radiologists manually drew the margin of the lungs, slice by slice, on CT images. The dataset used to develop the network consisted of 157 cases for training, 20 cases for development, and 26 cases for internal validation. Two-dimensional (2D) U-Net and three-dimensional (3D) U-Net models were used for the task. The network was trained to segment the lung parenchyma as a whole and segment the right and left lung separately. The University Hospitals of Geneva ILD dataset, which contained high-resolution CT images of ILD, was used for external validation.ResultsThe Dice similarity coefficients for internal validation were 99.6 ± 0.3% (2D U-Net whole lung model), 99.5 ± 0.3% (2D U-Net separate lung model), 99.4 ± 0.5% (3D U-Net whole lung model), and 99.4 ± 0.5% (3D U-Net separate lung model). The Dice similarity coefficients for the external validation dataset were 98.4 ± 1.0% (2D U-Net whole lung model) and 98.4 ± 1.0% (2D U-Net separate lung model). In 31 cases, where the extent of ILD was larger than 75% of the lung parenchymal area, the Dice similarity coefficients were 97.9 ± 1.3% (2D U-Net whole lung model) and 98.0 ± 1.2% (2D U-Net separate lung model).ConclusionThe deep neural network achieved excellent performance in automatically delineating the boundaries of lung parenchyma with extensive pathological conditions on non-contrast chest CT images.

Simultaneous attenuation and scatter correction of PET data in the image: quantitative and clinical assessment of image-to-image deep learning models.

The aim of this study is to analyze the application of early rehabilitation nursing in nursing intervention of neurological impairment among patients with acute ischemic stroke. 116 patients with acute ischemic stroke were selected as the research subjects in this paper. The patients were divided into 58 experimental (early rehabilitation care) and 58 control (routine rehabilitation care) groups according to the difference of care protocols, all of which were performed magnetic resonance imaging on. An image resolution reconstruction algorithm on the basis of deep convolutional neural network is proposed for MRI image processing. The results show that peak signal to noise ratio (PSNR) and structural similarity index measure (SSIM) of the included algorithm were remarkably greater than those of compressed sensing (CS) algorithm and nonlocal similarity and block low rank prior-based NSBL algorithm. Running time was shorter than that of the latter two algorithms (P < 0.05). The neurological impairment scores of patients in the experimental group 3 and 5 weeks after treatment were obviously lower than those of patients in the control group (P < 0.05). The Barthel indexes of patients in the experimental group 3 and 5 weeks after treatment were obviously higher than those of patients in the control group (P < 0.05). FugI-Meyer assessment (FMA) and Disability of Arm-Shoulder-Hand (DASH) scores of patients in the experimental group 3 and 5 weeks after treatment were obviously lower than those of patients in control group (P < 0.05). The results show that the deep learning algorithm for MRI image processing performance is better than the traditional algorithm. It not only improves the image quality but also improves the processing efficiency. Early rehabilitation nursing and routine rehabilitation nursing can effectively improve the neurological deficit symptoms, limb motor function, and daily living ability of patients with acute ischemic stroke, and the effect of early rehabilitation nursing is the best.

Super-resolution (SR) has received extensive attention in recent years for satellite image processing in a wide range of application scenarios, such as land classification, identification of changes, the discovery of resources, etc. Satellite images from satellite sensors are mostly low-resolution (LR) images, so they do not completely fulfill object detection and analysis criteria. SR has multiple residual network frameworks in deep learning that have improved performance and can extend thousands of layers in the system. However, each layer improves accuracy by doubling the number of layers, although training thousands of layers are too expensive, the process is slow, and there are functional recovery issues. We proposed a transferred wide residual Single Image Super-Resolution (SISR) remote sensing deep neural network model (WRSR). By increasing the width and reducing the residual network depth, the proposed approach has dramatically reduced memory costs. As a result, our model reduced memory costs by 21% in Enhanced Deep Residual Super-Resolution (EDSR) and 34% in SRResNet as a direct consequence of the in-depth reduction. The proposed architecture improves the efficiency of training loss by performing weight normalization instead of augmentation technology. We compared our method to five recent existing super-resolution (SR) deep neural network methods, tested over three public satellite image datasets and a standard reference (PRIM) dataset. Experiment analysis is evaluated in peak to signal noise ratio (PSNR) and structural similarity index measure (SSIM).