Search

Deep learning (DL) has shown great potential in conversions between various imaging modalities. Similarly, DL can be applied to synthesize a high-kV computed tomography (CT) image from its corresponding low-kV CT image. This indicates the feasibility of obtaining dual-energy CT (DECT) images without purchasing a DECT scanner. In this study, we investigated whether a low-to-high kV mapping was better than a high-to-low kV mapping. We used a U-Net model to perform conversions between different kV CT images. Moreover, we proposed a double U-Net model to improve the quality of original single-energy CT images. Ninety-eight patients who underwent brain DECT scans were used to train, validate, and test the proposed DL-based model. The results showed that the low-to-high kV conversion was better than the high-to-low kV conversion. In addition, the DL-based DECT images had better signal-to-noise ratios (SNRs) than the true (original) DECT images, but at the expense of a slight loss in spatial resolution. The mean CT number differences between the true and DL-based DECT images were within $$\pm$$ 1 HU. No statistically significant difference in CT number measurements was found between the true and DL-based DECT images (p > 0.05). The DL-based DECT images with improved SNR could produce low-noise virtual monoenergetic images. Our preliminary results indicate that DL has the potential to generate brain DECT images using single-energy brain CT images.

Object-based co-localization of fluorescent signals allows the assessment of interactions between two (or more) biological entities using spatial information. It relies on object identification with high accuracy to separate fluorescent signals from the background. Object detectors using convolutional neural networks (CNN) with annotated training samples could facilitate the process by detecting and counting fluorescent-labeled cells from fluorescence photomicrographs. However, datasets containing segmented annotations of colocalized cells are generally not available, and creating a new dataset with delineated masks is label-intensive. Also, the co-localization coefficient is often not used as a component during training with the CNN model. Yet, it may aid with localizing and detecting objects during training and testing. In this work, we propose to address these issues by using a quantification coefficient for co-localization called Manders overlapping coefficient (MOC)(1) as a single-layer branch in a CNN. Fully convolutional one-state (FCOS)(2) with a Resnet101 backbone served as the network to evaluate the effectiveness of the novel branch to assist with bounding box prediction. Training data were sourced from lab curated fluorescence images of neurons from the rat hippocampus, piriform cortex, somatosensory cortex, and amygdala. Results suggest that using modified FCOS with MOC outperformed the original FCOS model for accuracy in detecting fluorescence signals by 1.1% in mean average precision (mAP). The model could be downloaded from https://github.com/Alphafrey946/Colocalization-MOC.

PET images often suffer poor signal-to-noise ratio (SNR). Our objective is to improve the SNR of PET images using a deep neural network (DNN) model and MRI images without requiring any higher SNR PET images in training. METHODS: Our proposed DNN model consists of three modified U-Nets (3U-net). The PET training input data and targets were reconstructed using filtered-backprojection (FBP) and maximum likelihood expectation maximization (MLEM), respectively. FBP reconstruction was used because of its computational efficiency so that the trained network not only removes noise, but also accelerates image reconstruction. Digital brain phantoms downloaded from BrainWeb were used to evaluate the proposed method. Poisson noise was added into sinogram data to simulate a 6-minute brain PET scan. Attenuation effect was included and corrected before the image reconstruction. Extra Poisson noise was introduced to the training inputs to improve the network denoising capability. Three independent experiments were conducted to examine the reproducibility. A lesion was inserted into testing data to evaluate the impact of mismatched MRI information using the contrast-to-noise ratio (CNR). The negative impact on noise reduction was also studied when miscoregistration between PET and MRI images occurs. RESULTS: Compared with 1U-net trained with only PET images, training with PET/MRI decreased the mean squared error (MSE) by 31.3% and 34.0% for 1U-net and 3U-net, respectively. The MSE reduction is equivalent to an increase in the count level by 2.5 folds and 2.9 folds for 1U-net and 3U-net, respectively. Compared with the MLEM images, the lesion CNR was improved 2.7 folds and 1.4 folds for 1U-net and 3U-net, respectively. CONCLUSIONS: Our proposed method could improve the PET SNR without having higher SNR PET images.

Deep neural networks have been widely and successfully used in computer vision and attracted growing interests in medical imaging. In this work, we trained a deep residual convolutional neural network to improve quality of PET images. To train the deep neural network, we augmented real patient data with computer simulated phantom data. Specifically, we first trained the network using simulation data and then fine tuned the network using real data. Results based on simulation and real data show that the proposed method is more effective in removing noise than the traditional Gaussian filtering method.

PET scanners with partial-ring geometry have been proposed for various imaging purposes. The incomplete projection data obtained from this design cause undesirable artifacts in the reconstructed images. In this study, we investigated the performance of a deep learning (DL) based method for the recovery of partial-ring PET images. Twenty digital brain phantoms were used in the Monte Carlo simulation toolkit, SimSET, to simulate 15 min full-ring PET scans. Partial-ring PET data were generated from full-ring PET data by removing coincidence events that hit these specific detector blocks. A convolutional neural network based on the residual U-Net architecture was trained to predict full-ring data from partial-ring data in either the projection or image domain. The performance of the proposed DL-based method was evaluated by comparing with the PET images reconstructed using the full-ring projection data in terms of the mean squared error (MSE), structural similarity (SSIM) index and recovery coefficient (RC). The MSE results showed the superiority of the image-domain approach in reduction of 91.7% in contrast to 14.3% for the projection-domain approach. Therefore, the image-domain approach was used to study the influence of the number of detector block removal. The SSIM results were 0.998, 0.996 and 0.993 for 3, 5 and 7 detector block removals, respectively. The activity of gray and white matters could be fully recovered even with 7 detector block removal, while the RCs of two artificially inserted small lesions (3 pixels in diameter) in the testing data were 94%, 89% and 79% for 3, 5, and 7 detector block removals, respectively. Our simulation results suggest that DL has the potential to recover partial-ring PET images.

An intelligent travel experience management system, abbreviated as ITEMS, is proposed to help tourists organize and present the digital travel contents in an automatic and efficient manner. Readily available metadata are adopted to reduce the overhead of user intervention and manual annotation. Robust image similarity metrics are also incorporated to utilize the powerful searching capability of WWW search engines. The proposed system automatically identifies the inherent geo-information of personal media, and accordingly integrates media with map and text-based schedule to facilitate travel experience management and presentation. We show several prototype systems in two application scenarios and demonstrate the effectiveness of the proposed methodology.

Distributing video contents via broadcasting network mechanisms has become a promising business opportunity for the entertainment industry. However, since content piracy is always a serious problem, broadcasted contents must be adequately protected. Rather than implementing sophisticate key-management schemes for access control, an animation broadcast system based on colorization techniques is proposed. In the proposed system, gray-level animation video sequences are delivered via broadcast mechanisms, such as multicast, to reduce the overhead in server processing and network bandwidth. Moreover, color seeds labeled with fingerprint codes are delivered to each client through low-bandwidth auxiliary connections and then used to generate high-quality full-color animations with slight differences between versions received by each client-side device. When a user illegally duplicates and distributes the received video, his identity can be easily found out by examining features extracted from the pirated video. The proposed scheme also shows good resistance to collusion attacks where two or more users cooperate to generate an illegal copy in expectation of getting rid of legal responsibility. The proposed scheme exhibits advantages in network bandwidth, system performance and content security.

Purpose: To develop a diagnostic X-ray image simulator for commonly used out-patient-department examinations including lower extremities, skull, abdomen, pelvis, chest, and spinal cord. This simulator can be used as a computer-assistant-teaching software for training technologists more familiar with the relationship between image quality and operation conditions. Method and Materials: A detailed simulation of a diagnostic X-ray image requires four major inputs: operation conditions, X-ray spectral and spatial distributions, a human model, and cross-section corresponding to different materials. This study has collected commonly used operation conditions such as kVp, mAs, SSD for different OPD examinations. These conditions have been inputted to a Monte Carlo code, BEAMnrc, to build a virtual X-ray machine for calculating X-ray spectral and spatial distributions. After that, a high resolution voxelized human model, VIP-Man constructed from segmented Visible Male Dataset, will be imaged with this virtual X-ray machine. The cross-sections of different materials are generated by PEGS4, a cross-section preparation tool come with EGS4. Results: X-ray spectra simulated using BEAMnrc is almost identical to those listed in literatures, or from a spectra simulator, XCOMP3; except there is about 30% underestimation in characteristic peaks. However, this underestimation will lead to less than 0.1% deviation in our simulation. Heel effects can be observed in our simulated images. The intensity response was calibrated to a real digital X-ray machine, and point spread functions of this machine are measured to degrade our simulated images. Conclusion: We have developed this X-ray image simulator which can benefit the training of technologists. This simulator can be further improved to serve as a platform for studying image quality parameters such as QDE or MTF after we adding more realistic model of image receptor.

Iterative PET image reconstructions can improve quantitation accuracy by explicitly modeling photon-limited nature and physical effects of coincident photons. Geometric model in iterative reconstructions defines the mapping between image and sinogram domains based on scanner's geometry, which affects the accuracy of image results and the convergence rate of reconstruction algorithms. This paper examines the convergence rates of a reconstruction algorithm with three PET geometric models: interpolative, area-based, and solid-angle. The iterative algorithm used in this study is the maximum likelihood expectation-maximization (MLEM) algorithm. Experimental data are generated by the GATE package which simulates the Inveon microPET system. The comparison of convergence rate is based on the plot of log-likelihood value versus iteration number. From the plots of log-likelihood curves, the results from solid-angle model consistently reach the highest values at early iterations. It means that the MLEM algorithm with the solid-angle model will converge faster than the other two models. The experimental results indicate that the solid-angle model is a favorable geometric model for faster iterative PET image reconstruction.