The need for improved diagnostic methods in ophthalmology is acute, especially in the underdeveloped regions with limited access to specialists and advanced equipment. Therefore, we introduce VisionUnite, a novel vision-language foundation model for ophthalmology enhanced with clinical knowledge. VisionUnite has been pretrained on an extensive dataset comprising 1.24 million image-text pairs, and further refined using our proposed MMFundus dataset, which includes 296,379 high-quality fundus image-text pairs and 889,137 simulated doctor-patient dialogue instances. Our experiments indicate that VisionUnite outperforms existing generative foundation models such as GPT-4V and Gemini Pro. It also demonstrates diagnostic capabilities comparable to junior ophthalmologists. VisionUnite performs well in various clinical scenarios including open-ended multi-disease diagnosis, clinical explanation, and patient interaction, making it a highly versatile tool for initial ophthalmic disease screening. VisionUnite can also serve as an educational aid for junior ophthalmologists, accelerating their acquisition of knowledge regarding both common and underrepresented ophthalmic conditions. VisionUnite represents a significant advancement in ophthalmology, with broad implications for diagnostics, medical education, and understanding of disease mechanisms.

X-ray detectors in clinical computed tomography (CT) usually operate in current-integrating mode. Their complicated signal statistics often lead to intractable likelihood functions for practical use in model-based image reconstruction (MBIR). It is therefore desirable to design simplified statistical models without losing the essential factors. Depending on whether the CT transmission data are logarithmically transformed, pre-log and post-log models are two major categories of choices in CT MBIR. Both being approximations, it remains an open question whether one model can notably improve image quality over the other on real scanners. In this study, we develop and compare several pre-log and post-log MBIR algorithms under a unified framework. Their reconstruction accuracy based on simulation and clinical datasets are evaluated. The results show that pre-log MBIR can achieve notably better quantitative accuracy than post-log MBIR in ultra-low-dose CT, although in less extreme cases, post-log MBIR with handcrafted pre-processing remains a competitive alternative. Pre-log MBIR could play a growing role in emerging ultra-low-dose CT applications.

Dynamic (4D) CT imaging is used in a variety of applications, but the two major drawbacks of the technique are its increased radiation dose and longer reconstruction time. Here we present a statistical analysis of our previously proposed Mixed Confidence Estimation (MCE) method that addresses both these issues. This method, where framed iterative reconstruction is only performed on the dynamic regions of each frame while static regions are fixed across frames to a composite image, was proposed to reduce computation time. In this work, we generalize the previous method to describe any application where a portion of the image is known with higher confidence (static, composite, lower-frequency content, etc.) and a portion of the image is known with lower confidence (dynamic, targeted, etc). We show that by splitting the image space into higher and lower confidence components, MCE can lower the estimator variance in both regions compared to conventional reconstruction. We present a theoretical argument for this reduction in estimator variance and verify this argument with proof-of-principle simulations. We also propose a fast approximation of the variance of images reconstructed with MCE and confirm that this approximation is accurate compared to analytic calculations of and multi-realization image variance. This MCE method requires less computation time and provides reduced image variance for imaging scenarios where portions of the image are known with more certainty than others allowing for potentially reduced radiation dose and/or improved dynamic imaging.

Accurate system modeling in tomographic image reconstruction has been shown to reduce the spatial variance of resolution and improve quantitative accuracy. System modeling can be improved through analytic calculations, Monte Carlo simulations, and physical measurements. The purpose of this work is to improve clinical fully-3-D reconstruction without substantially increasing computation time. We present a practical method for measuring the detector blurring component of a whole-body positron emission tomography (PET) system to form an approximate system model for use with fully-3-D reconstruction. We employ Monte Carlo simulations to show that a non-collimated point source is acceptable for modeling the radial blurring present in a PET tomograph and we justify the use of a Na22 point source for collecting these measurements. We measure the system response on a whole-body scanner, simplify it to a 2-D function, and incorporate a parameterized version of this response into a modified fully-3-D OSEM algorithm. Empirical testing of the signal versus noise benefits reveal roughly a 15% improvement in spatial resolution and 10% improvement in contrast at matched image noise levels. Convergence analysis demonstrates improved resolution and contrast versus noise properties can be achieved with the proposed method with similar computation time as the conventional approach. Comparison of the measured spatially variant and invariant reconstruction revealed similar performance with conventional image metrics. Edge artifacts, which are a common artifact of resolution-modeled reconstruction methods, were less apparent in the spatially variant method than in the invariant method. With the proposed and other resolution-modeled reconstruction methods, edge artifacts need to be studied in more detail to determine the optimal tradeoff of resolution/contrast enhancement and edge fidelity.

Dual-energy (DE) X-ray computed tomography (CT) has been found useful in various applications. In medical imaging, one promising application is using low-dose DECT for attenuation correction in positron emission tomography (PET). Existing approaches to sinogram material decomposition ignore noise characteristics and are based on logarithmic transforms, producing noisy component sinogram estimates for low-dose DECT. In this paper, we propose two novel sinogram restoration methods based on statistical models: penalized weighted least square (PWLS) and penalized likelihood (PL), yielding less noisy component sinogram estimates for low-dose DECT than classical methods. The proposed methods consequently provide more precise attenuation correction of the PET emission images than do previous methods for sinogram material decomposition with DECT. We report simulations that compare the proposed techniques and existing approaches.

We present an efficient rebinning algorithm for positron emission tomography (PET) systems with panel detectors. The rebinning algorithm is derived in the planogram coordinate system which is the native data format for PET systems with panel detectors and is the 3-D extension of the 2-D linogram transform developed by Edholm. Theoretical error bounds and numerical results are included.

We present a simulation study of the global count-rate performance of a positron emission tomography (PET) scanner with different levels of partial collimation to maximize the noise equivalent count rate for whole-body PET imaging. We achieve partial collimation by removing different numbers of septal rings from the standard 2-D septa set for the GE Advance PET scanner. System behavior is studied with a photon tracking simulation package, which we modify to enable the production of random coincidences. The simulations are validated with measured data taken in 2-D and fully 3-D acquisition mode on a GE Advance system using the National Electrical Manufacturers Association NU-2 count-rate phantom with two sets of annular sleeves to expand the diameter to 27 and 35 cm. For all diameters and in 2-D and fully 3-D mode, there is good agreement between measurements and simulations. All studies use the three phantom diameters to evaluate the effect of patient thickness for each amount of collimation. Optimized system parameters, such as maximum ring difference for single slice rebinning, are determined for the five partially collimated systems considered. The resulting global count rates for true, scattered, and random coincidences, the noise equivalent count (NEC) rates, and the scatter fractions for different levels of collimation are compared along with the results from the conventional 2-D and fully 3-D modes. Improved statistical data quality relative to both 2-D and fully 3-D data is found with the partially collimated systems, particularly when one-half or two-thirds of the septal rings are removed. An increase in NEC rates of as much as 50% is found for clinically relevant activities between 5-10 mCi (184-370 MBq). Scatter fractions for the partially collimated systems are intermediate between the 2-D and fully 3-D numbers. Many factors that affect image quality have not been considered in this paper. However, the significant increase in statistical data quality warrants further investigation of the impact of partial collimation on clinical whole-body PET imaging.

Appropriate application of spatially variant system models can correct for degraded resolution response and mispositioning errors. This paper explores the detector blurring component of the system model for a whole body positron emission tomography (PET) system and extends this factor into a more general system response function to account for other system effects including the influence of Fourier rebinning (FORE). We model the system response function as a three-dimensional (3-D) function that blurs in the radial and axial dimension and is spatially variant in radial location. This function is derived from Monte Carlo simulations and incorporates inter-crystal scatter, crystal penetration, and the blurring due to the FORE algorithm. The improved system model is applied in a modified ordered subsets expectation maximization (OSEM) algorithm to reconstruct images from rebinned, fully 3-D PET data. The proposed method effectively removes the spatial variance in the resolution response, as shown in simulations of point sources. Furthermore, simulation and measured studies show the proposed method improves quantitative accuracy with a reduction in tumor bias compared to conventional OSEM on the order of 10%-30% depending on tumor size and smoothing parameter.

We present a method of performing fast and accurate three-dimensional (3-D) backprojection using only Fourier transform operations for line-integral data acquired by planar detector arrays in positron emission tomography. This approach is a 3-D extension of the two-dimensional (2-D) linogram technique of Edholm. By using a special choice of parameters to index a line of response (LOR) for a pair of planar detectors, rather than the conventional parameters used to index a LOR for a circular tomograph, all the LORs passing through a point in the field of view (FOV) lie on a 2-D plane in the four-dimensional (4-D) data space. Thus, backprojection of all the LORs passing through a point in the FOV corresponds to integration of a 2-D plane through the 4-D "planogram." The key step is that the integration along a set of parallel 2-D planes through the planogram, that is, backprojection of a plane of points, can be replaced by a 2-D section through the origin of the 4-D Fourier transform of the data. Backprojection can be performed as a sequence of Fourier transform operations, for faster implementation. In addition, we derive the central-section theorem for planogram format data, and also derive a reconstruction filter for both backprojection-filtering and filtered-backprojection reconstruction algorithms. With software-based Fourier transform calculations we provide preliminary comparisons of planogram backprojection to standard 3-D backprojection and demonstrate a reduction in computation time by a factor of approximately 15.

The combination of Fourier rebinning (FORE) and the ordered subsets expectation-maximization (OSEM), a fast statistical algorithm, appears as a promising alternative to the fully three-dimensional (3-D) iterative approach for clinical positron emission tomography (PET) data. In this paper, we evaluated the properties of FORE+OSEM and compared it with fully 3-D OSEM using both simulations and data acquired by commercial scanners. The aim is to determine to what extent the speed advantage of FORE+OSEM is paid for by a possible degradation of image quality in the case of noisy clinical PET data. A forward- and back-projection pair based on a line integral model was used in two-dimensional OSEM and 3-D OSEM (3D-OSEM) instead of a system matrix. Different variants of both approaches have been studied with simulations in terms of contrast-noise tradeoff. Two variants--FORE+OSEM with attenuation weighting (AW) [FORE+OSEM(AW)] and 3D-OSEM with attenuation-normalization weighting (ANSP) and a shifted-Poisson (SP) model [3D-OSEM(ANSP)]--were compared with measured phantom data and patient data. Based on the results from both simulations and measured data, we conclude that: 1) both attenuation (-normalization) weighting and the SP model improve the image quality but slow down the convergence and 2) despite its approximate nature, FORE+OSEM does not show apparent image degradation compared with 3D-OSEM for data with a noise level typical of a whole-body FDG scan.

The high computational cost of data processing in volume PET imaging is still hindering the routine application of this successful technique, especially in the case of dynamic studies. This paper describes two new algorithms based on an exact rebinning equation, which can be applied to accelerate the processing of three-dimensional (3-D) PET data. The first algorithm, FOREPROJ, is a fast-forward projection algorithm that allows calculation of the 3-D attenuation correction factors (ACF's) directly from a two-dimensional (2-D) transmission scan, without first reconstructing the attenuation map and then performing a 3-D forward projection. The use of FOREPROJ speeds up the estimation of the 3-D ACF's by more than a factor five. The second algorithm, FOREX, is a rebinning algorithm that is also more than five times faster, compared to the standard reprojection algorithm (3DRP) and does not suffer from the image distortions generated by the even faster approximate Fourier rebinning (FORE) method at large axial apertures. However, FOREX is probably not required by most existing scanners, as the axial apertures are not large enough to show improvements over FORE with clinical data. Both algorithms have been implemented and applied to data simulated for a scanner with a large axial aperture (30 degrees), and also to data acquired with the ECAT HR and the ECAT HR+ scanners. Results demonstrate the excellent accuracy achieved by these algorithms and the important speedup when the sinogram sizes are powers of two.

This paper presents two new rebinning algorithms for the reconstruction of three-dimensional (3-D) positron emission tomography (PET) data. A rebinning algorithm is one that first sorts the 3-D data into an ordinary two-dimensional (2-D) data set containing one sinogram for each transaxial slice to be reconstructed; the 3-D image is then recovered by applying to each slice a 2-D reconstruction method such as filtered-backprojection. This approach allows a significant speedup of 3-D reconstruction, which is particularly useful for applications involving dynamic acquisitions or whole-body imaging. The first new algorithm is obtained by discretizing an exact analytical inversion formula. The second algorithm, called the Fourier rebinning algorithm (FORE), is approximate but allows an efficient implementation based on taking 2-D Fourier transforms of the data. This second algorithm was implemented and applied to data acquired with the new generation of PET systems and also to simulated data for a scanner with an 18 degrees axial aperture. The reconstructed images were compared to those obtained with the 3-D reprojection algorithm (3DRP) which is the standard "exact" 3-D filtered-backprojection method. Results demonstrate that FORE provides a reliable alternative to 3DRP, while at the same time achieving an order of magnitude reduction in processing time.

The performance of the PENN-PET 240H scanner from UGM Medical Systems is tested and compared to the prototype PENN-PET scanner built at the University of Pennsylvania. The UGM PENN-PET scanner consists of six continuous position-sensitive NaI(Tl) detectors, which results in a 50 cm transverse field-of-view and a 12.8 cm axial field-of-view. The fine spatial sampling in the axial direction allows the data to be sorted into as many as 64 transverse planes, each 2 mm thick. A large axial acceptance angle, without interplane septa, results in a high-sensitivity and low-randoms fraction, with a low-scatter fraction due to the use of a narrow photopeak energy window. This work emphasizes those performance measurements that illustrate the special characteristics of a volume imaging scanner and how they change as the axial length is increased.