Full-Dose PET Image Estimation from Low-Dose PET Image Using Deep Learning: a Pilot Study.

Positron emission tomography (PET) imaging is an effective tool used in determining disease stage and lesion malignancy; however, radiation exposure to patients and technicians during PET scans continues to draw concern. One way to minimize radiation exposure is to reduce the dose of radioactive tracer administered in order to obtain the scan. Yet, low-dose images are inherently noisy and have poor image quality making them difficult to read. This paper proposes the use of a deep learning model that takes specific image features into account in the loss function to denoise low-dose PET image slices and estimate their full-dose image quality equivalent. Testing on low-dose image slices indicates a significant improvement in image quality that is comparable to the ground truth full-dose image slices. Additionally, this approach can lower the cost of conducting a PET scan since less radioactive material is required per scan, which may promote the usage of PET scans for medical diagnosis.

Automatic patient table removal in CT images.

In many medical imaging applications, it is desirable and important to localize and remove the patient table from CT images. However, existing methods often require user interactions to define the table and sometimes make inaccurate assumptions about the table shape. Due to different patient table designs, shapes, and characteristics, these methods are not robust in identifying and removing the patient table. This paper proposes a new automatic approach which first identifies and locates the patient table in the sagittal planes and then removes it from the axial planes. The method has been tested successfully against different tables in different products from multiple vendors, showing it is both a versatile and robust technique for patient table removal.

Pixel-feature hybrid fusion for PET/CT images.

Color blending is a popular display method for functional and anatomic image fusion. The underlay image is typically displayed in grayscale, and the overlay image is displayed in pseudo colors. This pixel-level fusion provides too much information for reviewers to analyze quickly and effectively and clutters the display. To improve the fusion image reviewing speed and reduce the information clutter, a pixel-feature hybrid fusion method is proposed and tested for PET/CT images. Segments of the colormap are selectively masked to have a few discrete colors, and pixels displayed in the masked colors are made transparent. The colormap thus creates a false contouring effect on overlay images and allows the underlay to show through to give contours an anatomic context. The PET standardized uptake value (SUV) is used to control where colormap segments are masked. Examples show that SUV features can be extracted and blended with CT image instantaneously for viewing and diagnosis, and the non-feature part of the PET image is transparent. The proposed pixel-feature hybrid fusion highlights PET SUV features on CT images and reduces display clutters. It is easy to implement and can be used as complementarily to existing pixel-level fusion methods.

Clinical applications of neurolinguistics in neurosurgery.

The protection of language function is one of the major challenges of brain surgery. Over the past century, neurosurgeons have attempted to seek the optimal strategy for the preoperative and intraoperative identification of language-related brain regions. Neurosurgeons have investigated the neural mechanism of language, developed neurolinguistics theory, and provided unique evidence to further understand the neural basis of language functions by using intraoperative cortical and subcortical electrical stimulation. With the emergence of modern neuroscience techniques and dramatic advances in language models over the last 25 years, novel language mapping methods have been applied in the neurosurgical practice to help neurosurgeons protect the brain and reduce morbidity. The rapid advancements in brain-computer interface have provided the perfect platform for the combination of neurosurgery and neurolinguistics. In this review, the history of neurolinguistics models, advancements in modern technology, role of neurosurgery in language mapping, and modern language mapping methods (including noninvasive neuroimaging techniques and invasive cortical electroencephalogram) are presented.

Fast PET Preview Image Reconstruction, Streaming, and Visualization During Data Acquisition: A Preliminary Study.

PET acquisition and reconstruction are time-consuming. A PET preview image is commonly reconstructed at the end of data acquisition of each bed-position frame in the step-and-shoot mode. We propose a scheme to reconstruct, stream, and visualize the PET preview image during acquisition to provide quasi-real-time visual feedback. Methods: As acquisition proceeds, event data are processed continuously by a backprojection method using time-of-flight kernels while corrections are applied only for sensitivity, time span, and decay. A preview update can be scheduled by frame or by a configured time interval. To create a preview image, the 3-dimensional volume of the current segment is knit with other existing segments. The knitted volume is projected onto a 2-dimensional plane, and the resultant gray-scale image is streamed to a display component for visualization. Results: By using fast and simple reconstruction and correction, the described scheme balances processing speed and image quality to provide early and frequent visual feedback. Results show that the preview creation, streaming, and visualization time are shorter than the acquisition time for a typical whole-body study. Conclusion: Fast feedback is achieved during PET acquisition, which provides clinicians with an indication of data acquisition and an estimation of image quality and allows early corrective measure and image quality control if necessary.

Ordered subset expectation maximization algorithm for positron emission tomographic image reconstruction using belief kernels.

The aim of this study is to investigate the benefits of incorporating prior information in list mode, time-of-flight (TOF) positron emission tomography (PET) image reconstruction using the ordered subset expectation maximization (OSEM) algorithm. This investigation consists of an IEC phantom study and a patient study. For the image under reconstruction, the activity profile along a line of response is treated as a priori and is combined with the TOF measurement to define a belief kernel used for forward and backward projections during the OSEM image reconstruction. Activity profiles are smoothed and combined with the TOF kernels to control the adverse impact of noise, and different levels of smoothness are attempted. The standard TOF OSEM reconstruction is used as a baseline for comparison. Image quality is assessed using a combination of visual assessment and quantitative measurement including contrast recovery coefficients (CRC) and background variability. On the IEC phantom study, the reconstruction using belief kernels converges faster and the reconstructed images are more appealing. The CRCs for all sizes of regions of interest on images reconstructed with belief kernels are higher than those of the baseline. The background variability, measured as a coefficient of variation, is generally lower for the images reconstructed using belief kernels. Similar observations occur on the patient study. Particularly, the images reconstructed using belief kernels have better defined lesions, improved contrast, and reduced background noise. OSEM PET image reconstruction using belief kernels that combine the information from prior images and TOF measurements seems promising and worth further investigation.

A Java program for stereo retinal image visualization.

Stereo imaging of the optic-disc is a gold standard examination of glaucoma, and progression of glaucoma can be detected from temporal stereo images. A Java-based software system is reported here which automatically aligns the left and right stereo retinal images and presents the aligned images side by side, along with the anaglyph computed from the aligned images. Moreover, the disparity between two aligned images is computed and used as the depth cue to render the optic-disc images, which can be interactively edited, panned, zoomed, rotated, and animated, allowing one to examine the surface of the optic-nerve head from different view angles. Measurement including length, area, and volume of regions of interest can also be performed interactively.

Mutual information-based registration of temporal and stereo retinal images using constrained optimization.

In the registration of temporal and stereo retinal images, the rotation angle is normally less than 5 degrees and the scaling factor is between 0.95 and 1.05. Due to sitting constraints in the imaging process, the x translation can be more than 100 pixels, but the y translation is usually small. This paper successfully incorporates these constraints in the mutual information-based registration and exploits a constrained optimization to seek an optimal registration. The proposed approach increases the success rate of the registration algorithm significantly. The impacts of the dynamic ranges of registration parameters on the registration outcome are studied and the effects of the order of rotation, scaling, and translation are also investigated.

An object-oriented framework for medical image registration, fusion, and visualization.

An object-oriented framework for image registration, fusion, and visualization was developed based on the classic model-view-controller paradigm. The framework employs many design patterns to facilitate legacy code reuse, manage software complexity, and enhance the maintainability and portability of the framework. Three sample applications built a-top of this framework are illustrated to show the effectiveness of this framework: the first one is for volume image grouping and re-sampling, the second one is for 2D registration and fusion, and the last one is for visualization of single images as well as registered volume images.

Influence of implementation parameters on registration of MR and SPECT brain images by maximization of mutual information.

UNLABELLED: Mutual-information maximization is one of the most popular algorithms for automatic image registration. However, many implementation issues have not been evaluated in a single, coherent context. METHODS: Twenty-one registrations between MR and SPECT brain images (8 patients) were achieved by mutual-information maximization with different implementation strategies. The results of a popular strategy were chosen as the standard. All other results were compared with the standard, and the statistics of misregistrations were computed. The registration speed, accuracy, precision, and success rate were assessed. RESULTS: Compared with trilinear interpolation, nearest-neighbor interpolation slightly sped the registration process, but with a lower success rate. The number of bins used to estimate the probability density function (pdf) affects the speed and robustness. Using fewer bins yielded a less robust registration. Adaptively changing the number of bins increased the registration speed and robustness. Simplex optimization increased the registration speed considerably, with a slightly degraded success rate. Simplex optimization with adaptive bin strategy improved the success rate and further decreased the registration time. Multiresolution optimization yielded a better success rate, with little effect on the accuracy and precision of registration. An increase in the number of resolution levels increased the success rate. Multisampling optimization also improved the success rate, but the results were less accurate and precise than those obtained with multiresolution optimization, with an increase in the number of levels decreasing the performance. Segmentation affected the registration speed and success rate. Because segmentation is problem specific, the effects were not conclusive. CONCLUSION: Different implementation strategies considerably affect the performance of automatic image registration by mutual-information maximization. On the basis of the experimental findings, we suggest that the best implementation strategy would include trilinear interpolation, adaptive change of the number of bins when estimating pdf, and exploitation of a simplex optimization algorithm with a multiresolution scheme.

Volume image registration by cross-entropy optimization.

Cross-entropy (CE), an information-theoretic measure, quantifies the difference between two probability density functions. This measure is applied to volume image registration. When a good prior estimation of the joint distribution of the voxel values of two images in registration is available, the CE can be minimized to find an optimal registration. If such a prior estimation is not available, one seeks the registration which gives a joint distribution different from unlikely ones as much as possible, i.e., the CE is maximized to find an optimal registration. When the unlikely distribution is a uniform one, CE maximization reduces to joint entropy minimization; when the unlikely distribution is proportional to one of the marginal distributions, it reduces to conditional entropy minimization; when the unlikely distribution is the product of two marginal distributions, it degenerates to mutual-information maximization. These different CEs are added together and are used as criteria for image registration. The accuracy and robustness of this new approach are tested and compared using a likely joint distribution and various unlikely joint distributions and their combinations.

Likelihood maximization approach to image registration.

A likelihood maximization approach to image registration is developed in this paper. It is assumed that the voxel values in two images in registration are probabilistically related. The principle of maximum likelihood is then exploited to find the optimal registration: the likelihood that given image f, one has image g and given image g, one has image f is optimized with respect to registration parameters. All voxel pairs in the overlapping volume or a portion of it can be used to compute the likelihood. A knowledge-based method and a self-consistent technique are proposed to obtain the probability relation. In the knowledge-based method, prior knowledge of the distribution of voxel pairs in two registered images is assumed, while such knowledge is not required in the self-consistent method. The accuracy and robustness of the likelihood maximization approach is validated by single modality registration of single photon emission computed tomographic (SPECT) images and magnetic resonance (MR) images and by multimodality registration (MR/SPECT). The results demonstrate that the performance of the likelihood maximization approach is comparable to that of the mutual information maximization technique. Finally the relationship between the likelihood approach and the entropy, conditional entropy, and mutual information approaches is discussed.

Volume sweeping and bodyline matching for automated prealignment in volumetric medical image registration.

For an automated image registration to converge to a good registration, it is crucial that the initial registration is within the capture range of the true registration, as local optimization methods are frequently employed. The ways to set an initial registration in current practice are not ideal and it is highly desirable to automate this initial registration (prealignment). Two automatic prealignment methods are reported here. In the volume sweeping approach, one volume is swept through the other, the overlapping volumes are coarsely aligned in the x and y directions, and a similarity measure is calculated at each sweeping position. Once sweeping is done, the position that gives the best similarity measure is chosen as the prealignment. In the second bodyline matching approach, patient bodyline profiles (the furthest anterior or posterior body boundary points) are extracted from two volumes and objectively matched. A prealignment is then derived from the matched bodylines. Both methods are tested on 19 PET/CT alignments of five patients with known ground truths acquired on hybrid PET/CT scanners. The absolute differences in the three translational parameters between the volume sweeping prealignment and the ground truth are 6.1 ± 3.9, 2.2 ± 2.7, and 4.2 ± 6.1mm, and between the bodyline matching and the ground truth are 5.2 ± 3.0, 3.3 ± 3.0, and 3.3 ± 4.1mm, which are within the capture range of automatic registration algorithms. The volume sweeping and bodyline matching can thus be used as a preprocessing step for automatic registration, making it possible to run an automatic registration algorithm without user intervention.