Multi‐angle acquisition and 3D composite reconstruction for organ‐targeted PET using planar detectors

Abstract

Background

This study investigates a multi‐angle acquisition method aimed at improving image quality in organ‐targeted PET detectors with planar detector heads. Organ‐targeted PET technologies have emerged to address limitations of conventional whole‐body PET/CT systems, such as restricted axial field‐of‐view (AFOV), limited spatial resolution, and high radiation exposure associated with PET procedures. The AFOV in organ‐targeted PET can be adjusted to the organ of interest, minimizing unwanted signals from other parts of the body, thus improving signal collection efficiency and reducing the dose of administered radiotracer. However, while planar detector PET technology allows for quasi‐3D image reconstruction due to the separation between detector heads, it suffers from degraded axial spatial resolution and, consequently, reduced recovery coefficients (RCs) along the axial direction perpendicular to the detectors.

Purpose

The purpose of this study was to evaluate the concept of multi‐angle image acquisition with two planar PET detectors and composite full 3D image reconstruction. This leverages data collection from multiple polar angles to improve the axial spatial resolution in the direction perpendicular to the detector heads. In such, the concept allows to overcome the intrinsic limitations of planar detectors in axial resolution.

Methods

This study evaluates the improvement in the quality of images acquired with the Radialis organ‐targeted PET camera through multi‐angle image acquisition, in both experimental and simulated imaging scenarios. This includes the use of custom‐made phantom with fillable spherical hot inserts, the NEMA NU4‐2008 image quality (IQ) phantom, and simulations with a digital brain phantom. The analysis involves the comparison of line profiles drawn through the spherical hot inserts, image uniformity, RCs, and the reduction of smearing observed in the axial planes with and without the multi‐angle acquisition strategy.

Results

Significant improvements were observed in reducing smearing, enhancing image uniformity, and increasing RCs using the evaluated multi‐angle acquisition method. In the composite images, the hot spheres appear more symmetrical in all planes. The image uniformity, calculated from the IQ phantom, improves from 7.79% and 10.98%, as measured in the images from the individual acquisitions, to 2.72% in the composite image. There is also an overall improvement in the RCs as measured from the hot rods of the IQ phantom. Furthermore, the simulation study using the digital human brain phantom demonstrates minimal smearing in the four‐angle scan, as opposed to a two‐angle scan.

Conclusion

The multi‐angle acquisition method offers a promising approach to transform planar PET detector technology into a true tomographic organ‐targeted PET system and to enable improvement in image quality while preserving a versatility inherent to planar detector technology. Future research will focus on optimizing the multi‐angle imaging protocol, including adjustments to detector separations, number of acquisition angles, and reconstruction iterations, alongside incorporating TOF, and reconstruction with point spread function modeling to further improve image quality.

1. BACKGROUND

The current trend in PET (positron emission tomography) involves tailoring its technology to suit the specific requirements of the imaging task at hand, including the emerging concept of total‐body (TB) PET. TB‐PET imaging represents a significant advancement in PET technology, expanding the axial field‐of‐view (AFOV) to encompass the entire patient body during image acquisition. ¹ , ² , ³ , ⁴ TB‐PET systems with an axial length in a range from 1.2 to 2 m maximizes the solid angle coverage, thereby boosting the detection efficiency of coincidence lines of response (LORs). This results in up to 40‐fold more events collected versus the whole‐body PET/computed tomography (WB‐PET/CT) scanner. ³ , ⁴ Consequently, TB‐PET allows for a substantial reduction in the injected dose of a radiotracer and/or acquisition time, minimizing patient motion and discomfort while improving throughput. However, TB‐PET scanners come with a higher cost due to up to tenfold increase in the number of individual detector elements compared to conventional WB‐PET/CT, posing a challenge to clinical adoption; compared to a 20 cm WB‐PET/CT detector, a TB‐PET detector with 100 and 200 cm axial lengths will result in 4 and 7.7 times higher component costs. ³ In addition, as the number of detector elements increases, challenges related to infrastructure and technology emerge, such as ensuring efficient data acquisition and storage, maintaining adequate cooling systems, and addressing issues with rapid image reconstruction. ³

It is important to note that there is a growing focus on clinical applications where imaging targets not the entire body but rather specific organs. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ In such cases, there is no need to surround the patient with extensive detector elements; instead, the emphasis is on enhancing the solid angle coverage of a targeted organ. This approach has the potential to yield superior image quality for specific organs while also offering the option to reduce radiation doses or shorten acquisition times—all at a reduced cost associated with PET imaging.

Organ‐targeted PET is considered for a wide range of applications, ¹⁰ including imaging of the breast, ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ prostate, ¹⁹ , ²⁰ , ²¹ , ²² pancreas ²³ as well as cardiac ²⁴ and neuroimaging. ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ To optimize imaging for specific organs, PET system designs vary from two‐planar detectors and ring geometries to more specialized “C”‐shaped and helmet‐like concepts. ¹⁰ , ³²

Nonetheless, there are two primary concepts for organ‐targeted PET scanners. One concept utilizes full‐ring detectors. Generally, ring detector architecture offers full 3D tomographic image acquisition with isotropic spatial resolution. ³³ These ring‐based PET detectors are implemented for organs such as the breast (e.g., MAMMI PET ¹⁷ ), brain (e.g., CareMiBrain, NeuroExplorer, VRAIN, ²⁸ MINDview, ²⁷ RF‐penetrable PET, ²⁶ PET‐Hat, ²⁵ 4D‐PET Brain System ³⁰ ), and prostate (e.g., ProsPET ²¹ ). The ring diameter and detector array configuration of each system are designed to maximize geometric efficiency for the specific organ under investigation. However, ring detector geometry restricts the system's versatility for applications beyond a targeted organ and results in a solo organ PET system with limited clinical utility. Furthermore, for breast cancer detection, ring detectors have reduced sensitivity in detecting lesions near the chest wall, imposing a risk that some lesions in this area may be missed. ³⁴ , ³⁵

An alternative concept employs two planar detectors positioned on either side of the organ of interest. ¹² , ¹³ , ¹⁴ , ¹⁶ , ¹⁸ , ¹⁹ With a certain (organ‐tailored) separation of detector heads during image acquisition, organ‐targeted PET can conduct limited‐angle tomographic scans and reconstruct quasi‐3D images. Unlike ring detectors, planar organ‐targeted detectors with adjustable separation and rotation provide flexibility in optimizing the AFOV for imaging various body parts. This adaptability significantly enhances the clinical utility of organ‐targeted PET systems. However, due to limitations in angular coverage, spatial resolution in the direction perpendicular to the detector heads deteriorates, affecting contrast recovery and overall image quality. ¹³ , ¹⁵ , ¹⁸ Therefore, while quasi‐3D images can be generated using iterative image reconstruction algorithms, providing high spatial resolution in the plane parallel to the detectors, ¹³ , ¹⁵ , ¹⁸ full 3D image reconstruction is necessary for accurate activity quantification, particularly in brain imaging, where isotropic spatial resolution is crucial for visualizing radiotracer uptake across sagittal, coronal, and axial planes.

The effective angular coverage of planar detectors can be increased by rotating the detectors during the scan. Systems developed by Smith et al. ¹¹ and Raylman et al. ¹⁴ and ClearPEM ¹² have utilized rotating detectors to improve the axial spatial resolution for breast imaging. Smith et al. investigated the spatial resolution, signal‐to‐noise ratio (SNR), and contrast in images acquired using both stationary and rotating detectors. They demonstrated that rotating detectors significantly improved image quality, SNR, and contrast compared to stationary acquisitions.

While previous studies have demonstrated the benefits of rotating planar detectors for breast imaging, ¹¹ , ¹² , ¹⁴ our approach extends this concept to create a versatile system capable of high‐quality imaging for both breast and brain applications, potentially expanding clinical utility of organ‐targeted PET technology and facilitating its clinical adoption.

In this study, we assess a method for 3D image reconstruction using data acquired at different angles (0°, ‐45°, +45°, +90°) with the Radialis PET camera, a high sensitivity low‐dose organ‐targeted PET scanner. This system comprises two planar detectors measuring 172 × 232 mm² similar to digital mammography systems. ¹⁸ Therefore, it is often referred to as positron emission mammography, or PEM, due to this resemblance. The high sensitivity of the Radialis PET camera is reflected in a significant dose reduction: while the standard clinical dose for 18F‐Fluorodeoxyglucose (18F‐FDG) radiotracer with WB‐PET is approximately 180–370 MBq, ³⁶ the recommended clinical dose for future practice using the current Radialis PET technology is only 74 MBq, ³⁷ representing up to 80% reduction compared to the standard dose.

For breast imaging with the Radialis PET camera, patients are scanned in a sitting position as schematically shown in Figure 1a. ³⁷ The detector heads are mounted on a rotating gantry, enabling the acquisition of images in craniocaudal (CC) and mediolateral oblique (MLO) views of each breast one at a time. The sensor area provides a sufficient AFOV to cover the entire breast during image acquisition (Figure 1a).

FIGURE 1.

(a) Configuration of the Radialis organ‐targeted PET with two planar detector heads. This rotation enables the acquisition of images of each breast in craniocaudal (CC) and mediolateral oblique (MLO) views one at a time. (b) The schematic of a brain image acquisition with the flexibility for the detector heads to be rotated from −90° to +90° around the head as well as at any angle in between.

Although the Radialis PET camera was first applied clinically to breast imaging, the versatility of its planar detector technology enables applications to other organs, such as the brain. The concept of brain imaging with the Radialis PET camera is schematically shown in Figure 1b. A larger detector separation is used for brain imaging compared to breast imaging. The rotating gantry provides the flexibility for the detector heads to be rotated around patient's head between −90° and +90°.

In this study, we use the Radialis PET camera to assess the image quality improvements achieved through rotating planar PET detectors. We conduct one‐ and two‐angle acquisitions of spherical fillable phantoms and a NEMA NU4‐2008 Image Quality (IQ) phantom. ³⁸ Our assessment includes evaluating the line profiles, image uniformity, and recovery coefficients (RCs) in all planes for the one‐ and two‐angle acquisitions. Additionally, we employ a digital brain phantom to assess the image quality achievable with a planar organ‐targeted PET detector using multi‐angle acquisition. Although time‐of‐flight (TOF) and reconstruction point spread function (PSF) modeling were not tested in this study, we anticipate that incorporating this information with multi‐angle acquisition will further improve image quality when planar PET detectors are used. ¹⁵ , ³⁹ , ⁴⁰

2. METHODS

2.1. System description

Planar detector heads of the Radialis organ‐targeted PET camera employ a 3 × 4 array of four‐side tileable sensor modules, ¹⁸ , ⁴¹ each containing a Cerium‐doped Lutetium Yttrium Orthosilicate (LYSO) scintillation crystal (pixelated to make a 24 × 24 grid with an individual pixel size of 2.32 × 2.32 × 13 mm³) and an array of 8 × 8 silicon photomultipliers (SiPMs) of the Array‐C type developed by ON Semiconductor (Phoenix, Arizona, USA). The LYSO crystals are optically coupled to the SiPMs via an uncoated 5 mm thick borosilicate light guide. In previous studies, we evaluated the performance of the Radialis PET camera using standardized and custom tests. We determined that this device has an in‐plane (transaxial) spatial resolution of 2.3 ± 0.1 mm and a cross‐plane (axial) resolution of 6.8 ± 0.1 mm with a detector separation of 80 mm. ¹⁸ The peak absolute sensitivity is 10.2% and average system sensitivity is 6.5%.

The detector heads are mounted on a gantry, which provides adjustable separation and rotation. In typical multi‐angle acquisitions, the gantry rotates the detector heads in a step‐and‐shoot manner. However, for this study, we used the Radialis PET detector heads independently of the movable gantry and rotated them manually.

Note that the acquisitions performed in this study were event‐based and not time‐based. The detectors were set to acquire a specified number of data records.

2.2. Image reconstruction

The standard image reconstruction method in the Radialis PET camera is used to individually reconstruct list mode file (LMF) data from each orientation of the detector heads using a 3D Maximum Likelihood Expectation Maximization (MLEM) algorithm. ⁴² To align the LMF datasets from different detector orientations to a common coordinate system, we set the coordinate system of the 0° acquisition as the reference frame. Transformation matrices are calculated to align each image with the reference image acquired at 0° position. These matrices are then applied to their respective LMF datasets, bringing all data into the same coordinate system as the 0° acquisition. The transformed LMF datasets from all detector orientations are then combined into a single dataset and used for the final composite 3D MLEM image reconstruction as shown in Figure 2.

FIGURE 2.

Schematic of the multi‐angle MLEM image reconstruction method.

To evaluate improvement in image quality following multi‐angle acquisitions and 3D composite image reconstruction, we also reconstructed the quasi‐3D images for each detector orientation individually, using the standard image reconstruction method. ¹⁸ , ⁴¹

The number of MLEM iterations was set to 15 for all reconstructions in this study. Normalization correction ⁴³ , ⁴⁴ and a median root prior filter ⁴⁵ are applied within the MLEM reconstruction after each iteration. It should be noted that a LOR angle‐allowance filter is implemented to reject events within the list‐mode data based on the endpoints of each LOR. Attenuation correction was not performed in this study. No Gaussian blurring or post‐processing corrections were applied during data analysis.

2.3. Phantom imaging

A custom‐made phantom utilizing a container with separate fillable spheres was used to assess improvements in lesion size estimation with the multi‐angle acquisition method. The green custom‐made phantom housing the spheres has a diameter of 120 mm and height of 75 mm. The fillable spheres have inner diameters of 5, 6.2, 7.9, and 12.4 mm, as shown in Figure 3a. The spheres were filled with 0.97 MBq/mL activity concentrations of 18F water and the phantom was placed in the FOV, as shown in Figure 3b. Then, a two‐angle acquisition was performed: first the detectors were oriented at 0°, parallel to the XY plane with 350 mm separation, and then rotated by 90° around the y‐axis to be parallel to the YZ plane with 258 mm separation, as illustrated in Figure 3b and c. The separation of 350 and 258 mm was chosen to mimic the same imaging conditions as would have been encountered when imaging a brain. The larger separation assumes imaging in posterior‐anterior directions with the addition of the head support beneath the patient's head. To use the same image reconstruction volume for both rotations, the number of slices parallel to the detector heads was chosen such to have a cubic image voxel with an edge size of 0.8 mm.

FIGURE 3.

(a) Custom‐made phantom with separate fillable hot spheres. (b) Illustration of the 0° acquisition of the spheres. The resolution will be degraded along the direction normal to the detectors, which is the z axis in the image coordinate system in this case. The high‐resolution (transaxial) plane will be formed by the X and Y axes of the 3D images. (c) Illustration of the 90° acquisition of the spheres. The resolution will be degraded along the x axis in the image and the transaxial plane will be formed by the z and y axes of the 3D images.

A standard NEMA NU4‐2008 IQ phantom was used to quantify improvements in the RCs with the multi‐angle acquisition method, Figure 4a. The IQ phantom consists of two regions: a solid part with five fillable rods of different diameters (5, 4, 3, 2, and 1 mm) to determine the activity RCs, and a fillable chamber to assess the image uniformity, using the analysis method we described in a previous study. ⁴¹ The absolute standard deviation for each RC value is calculated by determining the %STD_RC, as in the study by Baldassi et al., ⁴¹ and multiplying it by the RC value. The IQ phantom was filled with a 0.13 MBq/mL activity concentration of 18F water, oriented along the y axis in the FOV, and then the two‐angle acquisitions were performed as shown in Figure 4b and c. The detector separations used for imaging the IQ phantom were the same as for the custom‐made phantom described in the previous paragraph with 350 and 258 mm separations for 0° and 90° rotations, respectively.

FIGURE 4.

(a) Schematic of the NEMA NU‐4 Image Quality phantom with the side‐view (top) and a cross‐section view of the hot rods (bottom). ⁴⁶ (b) Illustration of the 0° acquisition of the IQ phantom. The resolution will be degraded along the direction normal to the detectors, which is the z axis in the image coordinate system in this case. The high‐resolution (transaxial) plane will be formed by the x and y axes of the 3D images. (c) Illustration of the 90° acquisition of the IQ phantom. The resolution will be degraded along the x axis in the image and the transaxial plane will be formed by the z and y axes of the 3D images.

Approximately 25 million events were acquired per angle in both experiments, namely, with the custom‐phantom with fillable spheres and the IQ phantom. The images from the 0° and 90° acquisitions were first reconstructed individually into quasi‐3D images using the standard image reconstruction method, illustrating the limitations in image quality without the multi angle acquisition. The data from both angles was then reconstructed together using the composite image reconstruction method, as described in Section 2.2. For all reconstructions, including both individual 0° and 90° acquisitions and the composite, the image space measured 258 × 216 × 350 mm³, comprising 322 × 216 × 438 voxels with cubic voxels of length 0.8 mm. The 3D images were resliced along the Z, X, and Y axes to assess the image quality in the XY, YZ, and XZ planes, respectively.

2.4. Digital brain phantom imaging

The digital brain phantom is a high‐resolution 3D volume, with voxel size 0.4 × 0.4 × 0.4 mm³, and it models 18F‐FDG uptake within the brain. Belzunce et al. ⁴⁷ used a single 18F‐FDG image from a real patient PET brain scan as a reference, along with histological data and corresponding tissue classified volume from the BigBrain atlas, ⁴⁸ and the Hammersmith atlas, ⁴⁹ to estimate the radiotracer uptake with a high spatial resolution and heterogeneity within the tissues.

We developed a custom program which simulates a specified number of true coincidence events from the voxelized digital brain phantom. The program generates true coincidence events from a randomly selected set of voxels from the phantom. The probability of an event occurring from a given voxel is calculated based on the relative uptake of radiotracer within that voxel. The detector sensing area in the simulation was the same as in the Radialis PET camera. Neither the scintillation light transport nor the photosensor detection processes were simulated in the custom program.

Simulations were conducted with three sets of tomographic scans, comprising two, four, and six‐angle acquisitions, using a digital 18F‐FDG PET brain phantom and planar PET detectors similar to the Radialis PET detectors. Total of 10 million events were acquired for each set (as this number of events is expected to be acquired for low dose head scans with the Radialis organ‐targeted PET camera), with 5, 2.5, and 1.67 million events per angle for the two, four, and six‐angle acquisitions, respectively. The rotation angle increments for each set were 90°, 45°, and 30° for two, four, and six‐angle acquisitions, respectively. An average adult human head has a thickness of 194 mm and breath of 145 mm. ⁵⁰ For the simulation, we randomly chose a separation of 230 mm, although in practice the separation may range from 200 to 300 mm. As an example, the detector orientations and head positioning for the four‐angle scan are illustrated in Figure 5, with the patient's head facing upwards.

FIGURE 5.

Illustrations of the simulated 4‐angle acquisition of the digital brain phantom.

The simulated LMF datasets were subsequently processed using the composite image reconstruction, as described in Section 2.2, to generate composite images, consisting of 577 × 433 × 577 cubic voxels with a length of 0.4 mm.

3. RESULTS

3.1. Custom‐made phantom with four separate hot spheres

Figure 6 compares selected slices from the 3D images of the hot spheres reconstructed individually for the 0° and 90° scans (rows a‐b) and using the composite reconstruction method (row c). In the individually reconstructed images, we observe that the spheres are undistorted in slices lying on the transaxial planes, which are parallel to the detector heads. However, the spheres are smeared along the axis normal to the detector heads, which is the z and x axis for the 0° and 90° scans, respectively, as indicated by the dotted, orange, arrows in Figure 3b. The smearing is significantly reduced in the composite image, in which the spheres appear more symmetrical in all planes, Figure 6 (row c).

FIGURE 6.

Selected slices from the 3D images of the custom‐made phantom with four separate spheres. Rows (a‐b) correspond to images reconstructed individually for the 0° and 90° scans and row (c) corresponds to composite images generated using the multi‐angle reconstruction method. The first, second, and third columns show selected slices from the XY, YZ, and XZ planes of the 3D images, respectively. The left column shows all four spheres in the FOV, the second column shows the 6.2 mm and 12.4 spheres, and the third column shows the 5 mm and 7.9 mm spheres. The slices highlighted in red correspond to the high‐resolution planes of the detector in each orientation. The sphere inner diameters are provided in the images in the first row.

We also assessed the reconstructed images using line profiles drawn through the maximum intensity voxels of each hot sphere. Figure 7 shows normalized line profiles through all spheres along the x, y, and z axis.

FIGURE 7.

Line profiles drawn across the 12.4, 7.9, 6.2, and 5 mm spheres along all three axes in the 3D images obtained from 0°, 90°, and composite reconstructions. In each plot, the vertical axis corresponds to the normalized pixel intensity and the horizontal axis corresponds to distance, in mm, from the center of each sphere. Please, note that the deviation from the zero background observed in the line profile for the 5 mm sphere in the 90° reconstruction is caused by 1) smearing along the x axis in the 90° acquisition and spillover of counts from the large 12.4 mm sphere into the region surrounding the 5 mm sphere; and 2) the offset of the 12.4 mm sphere along the y axis relative to the 5 mm sphere. This effect is mitigated in the composite 3D image.

As expected, the line profiles are considerably broader along the z axis in the 0° acquisition compared to x and y axis due to degraded spatial resolution along the z axis, which is normal to the detectors. Similarly, the line profiles are broad along the x axis in the 90° acquisition compared to the y and z axis. The line profiles along the z and x axes become considerably narrower and sharper in the composite image, reflecting a more isotropic distribution of activity within the spheres.

3.2. Image quality phantom

Figure 8 shows selected slices from the individually reconstructed and composite images of the IQ phantom. The phantom details are well visualized in the slices parallel to the transaxial planes, which are XY and YZ for the 0° and 90° scans, respectively. As expected, we observe a degradation of spatial resolution along the axis that is perpendicular to the detector heads. This degradation is coupled with a spill‐over effect, ⁵¹ where counts from the hot rods spread into adjacent areas causing increase in signal of the surrounding regions. The result is a significant smearing in a direction perpendicular to the detector heads. Slice 85 in the XZ plane from the 90° reconstruction shows how spill‐over effect results in overlap of the 5 and 4 mm rods. There is also significant spill‐in of counts from the adjacent hot background into the cold chambers.

FIGURE 8.

Selected slices from the 3D images of the IQ phantom. Rows (a‐b) correspond to images reconstructed individually for the 0° and 90° scans and row (c) corresponds to composite images generated using the multi‐angle reconstruction method. The first column shows single slices from the XY plane of the 3D images, the second column shows single slices from the YZ plane, and the last two columns show single slices from the XZ plane. The slices highlighted in red correspond to the high‐resolution planes of the detector in each orientation.

As is evident from Figure 8 (row C), smearing is significantly reduced in the composite 3D image. This leads to improved uniformity, as seen in the slices lying in the XY and YZ planes, and to a reduction in both the apparent spill‐out of activity from the hot rods and the spill‐in of activity into the cold chambers, as seen in the slices lying in the XZ plane.

Figure 9 shows line profiles drawn through the cold chambers, from slice 125, of the individual and composite reconstructions. The line profiles are drawn to compare the contrast in activity inside and outside the cold chambers for the individual and composite reconstructions. The valleys in the line profile correspond to centers of the cold chambers (15 mm center‐to‐center), and the peaks correspond to regions in the hot background. In the composite reconstruction, there is an improved contrast in intensities of the peaks and valleys as compared to the individual reconstructions. This indicates improved image contrast between the cold chambers and the hot background and reduced spillover of activity.

FIGURE 9.

Normalized line profiles through the cold chambers in the 0°, 90°, and composite reconstructions. Example of the line profile drawn through the cold chambers in the composite image is also shown on the plot.

We observed a significant reduction in smearing in all slices of the composite image compared to the images reconstructed individually from each acquisition. The line profile, however, is only shown for a single slice that passes through the midpoint of the central axis of the cold chambers.

3.2.1. Quantitative image quality analysis

Uniformity derived as a standard deviation from the mean grey value in the uniform region of the IQ phantom is 7.79% and 10.98% for the 0° and 90° reconstructions, respectively. The standard deviation, of 2.72%, is significantly lower in the composite image which indicates improvement in image uniformity.

The RCs for the 0°, 90°, and composite reconstructions were measured in slices parallel to the XZ plane and are shown in Figure 10. There is variation in the RCs between the two 0° and 90° acquisitions. However, there is an overall improvement in RC values in composite images for the 5 , 4, and 3 mm rods.

FIGURE 10.

RCs calculated for the 0°, 90° and the composite reconstruction. The absolute standard deviation for the RC values is provided in the brackets.

3.3. Simulated brain imaging with digital human brain phantom

Figure 11 presents slices of the reconstructed composite images from simulated multi‐angle acquisitions of the digital human brain phantom. No significant geometric distortions are observed in the sagittal and coronal planes of the two‐angle scan. However, the axial (low‐resolution) detector plane in the two‐angle scan exhibits smearing in the composite image. This diagonal smearing occurs due to missing projections between the 0° and 90° scans. The smearing is significantly reduced in the four‐angle scan, which involves two additional rotations (45° and 135°) of the detectors to acquire the missing projections. Moreover, no significant qualitative difference was observed in the images between the four‐angle scan and the six‐angle scan.

FIGURE 11.

(a) Axial, coronal, and sagittal views of the digital brain phantom. ⁴⁷ (b‐d) selected slices from the 3D composite images from the simulated two‐, four‐, and six‐angle acquisitions of the phantom.

4. DISCUSSION

In this work, we performed a set of tests to evaluate the image quality following multi‐angle acquisition and 3D composite image reconstruction using the Radialis organ‐targeted PET camera with planar detector heads. In previous studies, we characterized its performance and determined that the camera has a significantly higher count rate performance compared to WB‐PET/CT systems. ¹⁸ Although single‐angle image acquisition allows quasi‐3D image reconstruction, the axial spatial resolution (i.e., in the direction perpendicular to the detector heads) degrades due to limited angular view—an inherent limitation of planar PET detectors. As the result, smearing of objects arises and spillover effect exacerbates, which is particularly evident in loss of resolvability of the hot rods of the IQ phantom, Figure 8. We observed significant improvements in overall image quality of both, a phantom featuring hot spheres and the IQ phantom, following a two‐angle acquisition. As this is evident from Figure 8, by acquiring data from two orthogonal angles, we significantly reduced this smearing effect and achieved an improved spatial resolution in the direction perpendicular to the detector planes.

An improved spatial resolution is also reflected in enhancements of RC values. For instance, we showed an increase in RC values from 0.41 (absolute STD 0.03) and 0.53 (absolute STD 0.07) for single‐angle acquisitions to 0.59 (absolute STD 0.02) for the composite image of the 5 mm rod.

It is important to note that there are limited organ‐specific PET systems available for direct comparison with our results: in fact, the organ‐targeted imaging approach is an emerging field in medical imaging technology. Despite this, we can draw some comparisons. For instance, Raylman et al. ¹⁴ reported the RCs for a rotating PET/CT scanner under development for breast cancer imaging. Their system achieved RCs of 0.89, 0.80, 0.69, 0.33, and 0.12 for the 5, 4, 3, 2, and 1 mm rods, respectively. In comparison, our composite images yielded RCs of 0.59, 0.43, and 0.25 for the 5, 4, and 3 mm rods, respectively. Meanwhile, the 2 and 1 mm rods were not visible in the images and the RC was not calculated for these rods. The discrepancy between the cited literature and our measurements may be attributed to several factors. First, the referenced PET/CT system rotates in nine steps of 20°. In our experiment, we performed two 90° rotations, which is not sufficient to achieve complete angular sampling for the detector separations used in the acquisitions. Additionally, we did not perform attenuation corrections in this study, while the referenced PET/CT system uses CT‐based attenuation correction that enhances the quantitative accuracy and directly impacts RCs. It should be noted that the CT component in the referenced system significantly increases radiation exposure, making it unsuitable for screening applications with undiagnosed patients. In contrast, Radialis PET camera is specifically designed for low‐dose applications including screening where limiting cumulative radiation dose is crucial. ¹⁸ , ³⁷ , ⁴¹ , ⁵² Nevertheless, the Radialis PET camera has the ability to perform attenuation correction through image segmentation, ⁵³ and we plan to evaluate this method with multi‐angle imaging in the future studies.

An additional advantage of the Radialis PET camera is its versatility, which enhances its clinical utility and cost‐effectiveness in clinical practice compared to systems dedicated to imaging a single organ (such as breast or brain). ¹⁷ , ²¹ , ²⁹ , ³¹ This system allows for dynamic adjustment of separation between planar detector heads and their easy positioning around different organs of interest. As well, the number of projections can be tuned, ensuring optimal angular coverage for various organs. This flexibility improves geometric sensitivity and enables low‐dose PET imaging of various body parts, including but not limited to the breast, brain, and potentially other organs like the prostate or heart.

Although the clinical utility of organ‐targeted PET with planar detectors extends beyond breast and neuroimaging, multi‐angle imaging can be readily adopted for clinical practice for these two applications. For breast imaging, the dual‐angle acquisitions will be performed with the patient in seated position as schematically shown in Figure 1a. To properly support the breast during image acquisition, and to address issues of varying breast deformation at different detector positions and patient motion, a specialized vacuum‐assisted breast immobilization holder will be utilized. ⁵⁴ Each breast will be immobilized and scanned one at a time. After immobilizing the breast, the detectors will be placed horizontally to acquire images in the CC view. They will then be lifted and adjusted to a vertical position for MLO view imaging. The images captured from two angles will be used to create composite 3D images. It should be noted that while single‐projection acquisitions (in CC and MLO view) have been shown to detect breast lesions with high sensitivity and specificity, ³⁷ , ⁵⁵ the dual‐angle acquisition can significantly enhance diagnostic potential. Indeed, the AFOV of the Radialis detector heads, coupled with their relatively close proximity (∼8 cm for an average breast size) during breast imaging, provides full angular coverage of breast. This improves RCs and facilitates accurate assessment of metabolic changes in small lesions, which is critical for precise monitoring of treatment response, where subtle metabolic changes can indicate therapeutic effectiveness.

In brain imaging (Figure 1b), the patients will lie supine on a bed with their head restrained using a headrest. The detectors rotate freely around the patient's head, covering an angular range from −90° to +90°. However, phantom experiments reveal that the larger inter‐detector separation required for brain imaging, compared to breast imaging, precludes complete angular coverage with the dual‐angle acquisition protocol. This leaves missing projections between the 0° and 90° acquisitions, which contribute to image distortion. Our simulations of a digital brain phantom demonstrated improved visualization of brain features and a reduction in smearing, especially following the four‐angle reconstruction. This improvement is crucial for visualizing fine functional details, which are essential for diagnosing complex neurological conditions. So, as a next step, we will perform phantom experiments with a greater number of detector rotations, to optimize the multi‐angle imaging protocol for various clinically relevant detector head separations, thereby enhancing the system's versatility and diagnostic capabilities.

It should be mentioned that given the high count sensitivity of the Radialis organ‐targeted PET camera, the acquisition time per angle is expected not to exceed 5 min for the lowest activity of an injected radiotracer. Thus, the multi‐angle acquisition approach is not expected to impact the patient's throughput.

Moving forward, our research team is exploring options for the use of TOF capable detectors to further improve spatial resolution using Application‐Specific Integrated Circuits (ASICs) based data acquisition systems. Literature suggests that timing resolution in the range of 230–275 ps is attainable with ASICs. ⁵⁶ Considering the fact that the best timing resolution obtained with detectors based on SiPM and LYSO crystals is about 100 ps ⁵⁷ (which translates to 3 cm in spatial domain), the implementation of TOF in organ targeted PET application, along with the multi‐angle imaging method, might be justified for imaging of organs where larger detector separations are used (e.g., for brain imaging where detector separation is 20–30 cm).

5. CONCLUSION

We have demonstrated that the multi‐angle acquisition technique with planar PET detectors offers a practical solution for achieving high‐resolution 3D tomographic imaging with planar organ‐targeted PET systems. Our findings showed significant enhancements in image quality, including improved image uniformity, contrast, and reduced image smearing through multi‐angle acquisition and 3D composite image reconstruction. This advancement is particularly promising for clinical applications requiring high spatial resolution and precise radiotracer uptake reconstruction, such as the accurate detection and characterization of neurological conditions in brain imaging, early stages of cancer, and quantifying activity changes for therapy assessment.

In future work, we plan to optimize the number of acquisition angles and reconstruction iterations to find an optimal balance between image quality and scanning time for different clinical applications. Additionally, integrating TOF reconstruction and PSF modeling into the image reconstruction process is expected to further enhance image resolution and contrast recovery.

CONFLICT OF INTEREST STATEMENT

AR and OB are co‐founder of Radialis Inc. They each own 7% of the outstanding shares in Radialis Inc. EA, BK, BB, AB and OB are employed by Radialis Inc.

Shahi A, Poladyan H, Anashkin E, et al. Multi‐angle acquisition and 3D composite reconstruction for organ‐targeted PET using planar detectors. Med Phys. 2025;52:2507–2519. 10.1002/mp.17606