Abstract
Extensive research is currently being conducted into dynamic positron emission tomography (PET) acquisitions (including dynamic whole-body imaging) as well as extraction of radiomic features from imaging modalities. We describe a new PET viewing software known as Imager-4D that provides a facile means of viewing and analyzing dynamic PET data and obtaining associated quantitative metrics including radiomic parameters. The Imager-4D was programmed in the Java language utilizing the FX extensions. It is executable on any system for which a Java w/FX compliant virtual machine is available. The software incorporates the ability to view and analyze dynamic data acquired with different types of dynamic protocols. For image display, the program maintains a built-in library of 62 different lookup tables with monochromatic and full-color distributions. The Imager-4D provides multiple display layouts and can display fused images. Multiple methods of volume-of-interest (VOI) selection are available. Dynamic analysis features, such as image summation and full Patlak analysis, are also available. The user interface includes window width and level, blending, and zoom functionality. VOI sizes are adjustable and data from VOIs can either be displayed numerically or graphically within the software or exported. An example case of a 50-year-old woman with metastatic colorectal cancer and thyroiditis is included and demonstrates the steps for a user to obtain standard PET parameters, dynamic data, and radiomic features using selected VOIs. The Imager-4D represents a novel PET viewer that allows the user to view dynamic PET data, to derive dynamic and radiomic parameters from that data, and to combine dynamic data with radiomics (“dynomics”). The Imager-4D is available as a free download. This software has the potential to speed the adoption of advanced analysis of dynamic PET data into routine clinical use.
Keywords: Dynamic PET, FDG PET, Radiomics, Dynomics
Introduction
The acquisition and interpretation of positron emission tomography (PET) data is growing progressively more complex with the advent of scanners capable of dynamic whole-body protocols [1–3] as well as the emergence of advanced quantitative metrics for lesion characterization and prognostication such as radiomics features [4, 5]. Information such as dynamic radiotracer uptake and radiomics may add significantly to the diagnostic workup of a patient [6–8], and growing evidence suggests the ability to influence treatment outcome predictions and personalize treatment for patients with lung cancer, oropharyngeal squamous cell carcinoma, bone and soft tissue lesions, esophageal cancer, glioma, lung cancer, and pancreatic lesions [9, 10]. PET viewing software packages deployed in the clinic are often not optimized for the display of dynamic images or for obtaining novel radiomic parameters. Indeed, while many such software packages are proprietary, the few reports in the literature on academic home-grown software for PET viewing have focused on the display of static, single-timepoint PET images and the extraction of more traditional parameters such as standardized uptake values (SUVs) [11]. So long as dynamic PET and radiomics are generally analyzed on expensive, specialized research workstations, there will be sizable barriers to their implementation in the clinic.
We therefore set out to significantly expand upon the standard capabilities of PET viewers through the development of software that would (1) cinematically display dynamically acquired data including such data acquired on conventional PET/CT or PET/MRI systems, (2) allow the user to place volumes-of-interest (VOIs) on the dynamic data at any timepoint and generate quantitative parameters from that data, (3) allow the user to extract radiomic parameters from either static or dynamic PET data, and (4) be freely available for download for any interested clinician-scientists in order to speed the adoption of dynamic and radiomic features into clinical workflow [12]. This new software is known as Imager-4D, and the present manuscript will describe its design and implementation as well as provide an interpretive case study on the use of this novel PET viewer.
Materials and Methods
Imager-4D General Principles and Considerations for Dynamic Image Viewing
The Imager-4D was developed to provide a platform for the display and analysis of dynamically acquired whole-body multi-modality co-registered image sets, in particular PET/CT and PET/MR. In addition to utilizing similar display and analysis functionality to that commonly available for single timepoint imaging, it extends the typical quantitative analysis by also providing integrated Patlak modeling and VOI-based radiomic analysis of all original and derived datasets. The software is intended to be as vendor-neutral as possible, with only standard Digital Imagine and Communication (DICOM) fields accessed for demographic information.
As a result of the nature of the data it was designed to utilize, the Imager-4D often requires large amounts of RAM memory (for static timepoint PET/CT display and analysis, it is recommended that the user provides a minimum of 4 GB of RAM to the system; for dynamic whole-body PET/CT, with Patlak analysis, it is not uncommon for 20 GB or more to be needed to fully analyze and visualize the results). Ultimately, the minimum recommended computational memory will, of course, depend on the number of dynamic PET frames and the total number of voxels per frame. In addition, the Imager-4D is able to utilize multi-core and multi-processor capabilities of the underlying machine, when available, by parallelizing many non-sequential processing tasks. For potentially time-consuming operations, the program will display a progress-bar informing the user of a task’s status. Due to the memory requirements, it is recommended, though not required, that any underlying platform as well as the Java VM be of a 64-bit variety. Execution on a 32-bit system may be satisfactory for single bed-frame position imaging, but whole-body dynamic datasets will generate intermediate and derivative datasets which will likely exceed the memory capabilities of a 32-bit machine.
The Imager-4D was written in the Java programming language version 8 utilizing the FX extensions (Oracle, Redwood Shores, CA, USA). As such, it is executable on any system for which a Java w/FX compliant virtual machine (VM) is available. Unlike single-timepoint datasets, dynamic data can be acquired and encoded in multiple ways which requires a flexible data loading mechanism that can restructure incoming data as required. The goal is a 4D dataset where the fourth dimension represents time and each element of the time array is a 3D static dataset. Some scanners acquire and encode the entire temporal dataset as a single series, embedding both spatial and temporal markers within each file. For these datasets, the program was designed to determine at the onset of data loading the final structure of a fully loaded dataset. For other systems, particularly those which do not possess native whole-body dynamic acquisition capabilities but instead utilize repeated, operator-initiated whole-body imaging sequences, the program must read all of the available image data first, adding image data to a self-organizing “bucket” which dynamically sorts data both temporally and spatially as it is loaded into the system. Only when all available data has been loaded can it review the resulting 4D structure and validate the completeness of the data.
As illustrated in Fig. 1, dynamically acquired whole-body (WB) PET studies consist of multiple whole-body passes acquired sequentially over the duration of imaging. On systems capable of native dynamic whole-body acquisition, the image data is typically reconstructed and saved as a single DICOM series, appropriately decay-correcting image data to the designated time (either acquisition start time or time of injection, depending on user settings). However, for systems without native support for dynamic WB acquisition, individual WB passes may be decay-corrected to their respective acquisition start time. As such, one must further decay-correct the image data to the start of the first WB pass or, depending on the normalization to be deployed (e.g., SUV), to the time of injection. The software incorporates the ability to differentiate between these types of acquisitions and automatically makes the appropriate calculations. For other imaging types, such as CT and MRI, the same data loading and normalization are performed, though transformations unnecessary for those modalities are ignored.
Fig. 1.
Schematic representation of dynamic PET data acquisition
Finally, transformation into user-selected units, such as SUV (and its variants), occurs through the use of a separate scalar value which is applied to the voxel value whenever the voxel is read from its source buffer. The scalar is not applied to the buffer value directly, however allowing the program to change units without regenerating the underlying normalized and decay-corrected image data.
Bitmap Generation, Lookup Tables, and Memory Management
The program maintains a 4D array of bitmap objects as buffers for the display bitmaps needed for each source dataset loaded, including derived datasets. These bitmaps are generated directly from the source data, utilizing user-selected lookup tables (LUTs) and user-specified window-level settings (Fig. 2), and are generated on an “as-needed” basis. They are retained until corrupted by either a user action (e.g., change of LUT and adjustment of window-level settings), or the system needs to reclaim memory from currently unused bitmaps for other tasks or data requirements.
Fig. 2.
Imager-4D user interface window. In this image, the window is configured to display CTS static-timepoint PET/CT from a 50-year-old patient with metastatic rectal cancer and thyroiditis. The LUT controls that allow user control of window W/L settings are on the left of the window (large red arrow for CT and red arrowhead for PET). The top control ribbon (medium red arrow) contains multiple controls including for layout, information display, and triangulation indicator. Zoom functionality is located along the right side of the image display (thin red arrow). Multiple VOIs are displayed on the CTS images including normal liver (black circle on the coronal image), liver metastases (blue and red circles on the coronal and axial images, respectively), and thyroid (green circle on the sagittal image)
The program maintains a built-in library of 62 different LUTs with various monochromatic and full-color distributions. All are encoded as 24-bit color and are of varying lengths, ranging from 256 to 768 unique values. Bits 25–32 of the color entry in each palette are reserved for use by the palette manager to control the blending of LUTs when the software is displaying fused datasets. The LUTs are built in to the LUT control UI object which appears adjacent to the image display window (Fig. 2). The LUT control uses as its primary user-interface object a simple rectangular bitmap of the lookup table itself. The top and bottom of the LUT control contain text fields which display the current minimum and maximum of the window-level (W/L) range selected. A “zoom” feature allows the user to limit the range of the LUT control, allowing the user to manipulate the LUT with higher resolution, whereas the “reset” feature allows restoration of control to the full dynamic range of the image dataset.
Layout Manager
The Imager-4D provides multiple display layouts: axial, sagittal, and coronal views (or all three simultaneously displayed), as well as the display of maximum intensity projection (MIP) images. The layout of the screen is set by a “layout” combo-box control on the top control ribbon (Fig. 2) in the application. Available settings include “axial,” “coronal,” “sagittal,” “TCS” (Transaxial, Coronal, Sagittal combined (Fig. 2)), and “MIP.”
Fusion Display
If a second, co-registered dataset is loaded into the software, it is co-displayed as a fused dataset in all of the available display layouts. When displaying fused datasets, the only change to the program is the availability within the user-interface of a second LUT tool for manipulation of the LUT for the secondary study (Fig. 2). The same features (e.g., color themes and controls) described for the single LUT control are also available for each control when dual modalities are displayed.
When the system is in fusion display mode, there also appears a blending tool between the LUT controls that changes the relative mixture of the two images within the fused display. Blending can range from 0%, representing the display of the primary dataset only, to 100%, which represents the display of the secondary dataset only. The default setting is 25%, which allows 25% saturation of the primary image (usually PET) with a 75% contribution by the secondary image (usually CT or MRI) to the final display.
Display Features
The correct display and analysis of any dataset may require additional normalization and transformation to correct for other modality-dependent factors, such as radioactive decay or injected activity, as in the case of PET imaging. The software allows for such corrections to be input by the user.
The software provides a typical set of display features, though the two that would likely be most commonly used are the (1) information display (
) and (2) triangulation indicator (
). These appear as buttons along the top control ribbon in the application window (Fig. 2). When the triangulation indicator is selected, a set of cross-hairs will appear in whichever image display windows are currently displayed (except for the MIP display) which will be centered on a global image space triangulation point. When the information display button is pressed, key demographic information derived mostly from the primary image set is displayed overlying the image. This information includes patient name, patient medical record number, study date and type, modality, and the date and time of acquisition. In addition, position information related to the view and global image space triangulation point will appear, as well as the normalized and transformed voxel value for the voxel currently selected by the cross-hairs for both displayed modalities.
The text and graphics parameters for these display elements, such as color and font, can be changed by selecting the “options-display” menu chain. From there, the color of the cross-hairs can be changed, as well as the font size, style, and color of the information display. In addition, if the user wishes to de-identify the data display, selecting the de-identify option will strip the patient’s name, medical record number, and study date from all display elements, as is demonstrated in the included figures.
Volumes-of-Interest
Image quantitation is made available through the use of interactively drawn spherical or cube-shaped volumes-of-interest (VOIs). The size of these VOIs is determined by a radius, which is consistently applied across all three axes; thus, spheres are always spheres and cubes are always symmetric cubes, regardless of the underlying voxel dimensionality. The minimum radius for a sphere is twice the smallest voxel dimension. The minimum width for a cube is the minimum odd number of voxels (> 1) with an aggregate dimensionality along the axis of the smallest size which is the nearest to the user-drawn size.
VOI Sampling
The user can select which VOI to draw by right-clicking the image surface and selecting the VOI type from the context menu that appears. The VOI then appears in its default size. The user can then translate the VOI through the volume by left-clicking the mouse within the VOI and, while keeping the mouse pressed, move the mouse through the image volume. The user can resize the VOI by holding the CONTROL key down on the keyboard while selecting the VOI with the mouse. While holding the mouse button down and translating the mouse to and from the center of the VOI, the user can resize the VOI keeping the center of the VOI fixed in its current position. The user can control certain features of a VOI, including assigning a name to the VOI and the drawing color for the VOI. The VOIs can be saved into a proprietary “.roi” file for later retrieval and use.
By default, sampling of the voxels defined by a VOI occurs on an event-driven basis, and occurs automatically as the user draws, resizes, or translates a VOI across image space. However, in the special case of dynamic datasets, the program will switch into a sampling mode where image sampling will be deferred until the user ceases to manipulate the VOI, at which point re-sampling proceeds automatically.
Sampling of VOIs provides multiple descriptive statistical measures, including maximum, minimum, peak (PET Only), mean, standard deviation, coefficient of variation, voxel count, and volume. Additionally, the program provides radiomic texture-based statistics, including skew, kurtosis, energy, entropy, homogeneity (both inverse difference moment (IDM) and angular second moment (ASM)), contrast, correlation, dissimilarity, autocorrelation, absolute value, and cluster (tendency, shade, and prominence). The Imager-4D also provides for manipulation of various parameters when calculating the texture-based statistics. These parameters include the use of a globally or locally generated gray-level co-occurrence map, the number of gray-levels of the co-occurrence map, the distance (in voxels) from the reference voxel for co-occurrence calculation, and additional settings which allow the user to fine-tune texture calculations for their specific needs. For PET data, the program also allows the auto-calculation of the number of gray-levels based on a user-defined step function based on the transformed image units of the image (e.g., binning based on SUV).
When VOIs are sampled on fused image sets, VOI statistics are generated for both the primary (usually functional) and secondary (usually anatomical) datasets. As the VOIs are spatially defined (real-world dimensions, not voxels), they are separately sampled to the underlying voxel dimensionality of each dataset before quantitative sampling takes place. Thus, the quantitation usually encompasses similar volumes for each dataset, although the number of voxels included in each calculation may vary greatly if the voxel dimensionality of each dataset similarly varies.
VOI Statistic Reporting
A unique feature of the Imager-4D is that all dynamic image processing is performed on a “progressive” basis, meaning that all steps of a sequential processing task are retained for review, quantitation, and evaluation. As an example, if a 15-timepoint study is loaded into the software and the user wants to generate a “summed” dataset from the source data, the “summed” dataset will also have 15 timepoints of progressively summed data, with the first timepoint of the summed data containing data from the first timepoint of source data only, and the last timepoint of the summed data containing the results of summing all timepoints of source data (with intermediate timepoints containing the summed data from the initial timepoint through that timepoint only).
Patlak Analysis
The software is capable of performing multiple types of processing on dynamic image data. In addition to the progressive summation just described, the software can perform a simple linear least squares fitting of voxel data as well as a linear Patlak analysis (Eq. 1) where an input function (which can be defined by the user using multiple methods) and the tissue dynamic activity concentration can be fitted linearly. For both fitted techniques, an associated “goodness-of-fit” image matrix is also generated, using the correlation coefficient resulting from the least squares fitting calculation:
| 1 |
The Patlak graphical analysis [13–15], as applied to PET imaging, assumes irreversible binding or uptake of radiotracer by the tissue and is described mathematically by Eq. 1 where R(t) reflects the radiotracer measured in an ROI (or voxel) at time t and Cp(t) reflects the radiotracer in plasma/blood at time t. The goal is to solve for “K” and “V0,” where “K” is the overall tracer influx rate by which the radiotracer accumulates in the irreversible compartment, and V0, sometimes referred to as the distribution volume, represents the unbound radiotracer.
When loaded with appropriate dynamic data, the program internally calculates a progressive Patlak plot on a voxel-by-voxel basis and provides for the display and VOI-based analysis of the progressive slope and intercept image sets calculated by the individual voxel Patlak plots at each timepoint beginning with the second timepoint of data (the least squares fitting routine requiring at least 2 datapoints to generate a result). In addition, the least squares fit component of the Patlak image generation also calculates a voxel-by-voxel correlation coefficient which provides a parameter for the “goodness-of-fit” for the modeling of that voxel over time. This goodness-of-fit voxel map can be used to filter voxels for display, allowing the user to limit the display of any dataset to only those voxels whose kinetics are correlated over a user-specified threshold (e.g., only display pixels with an r2 > =0.9) [16].
Results
The User Interface: How to Navigate Imager-4D
Once a dataset is loaded into the Imager-4D, and the validated internal data structures are created and populated, scalars are read from the individual image meta-headers, and the original voxel values are normalized to the encoded image units of the dataset.
The W/L range (Fig. 2) can be manipulated by clicking and holding the left mouse button on either the top or bottom of the bitmap and dragging it within the area of the control. If the user wishes to translate the selected W/L range, then clicking and holding the left mouse button within the middle of the LUT field and then dragging the mouse up or down will translate the range in the desired direction. The user is not allowed to adjust the range outside the natural range of values of the underlying dataset, but they can “zoom-in” to a small set of values by approximating the small range and then clicking the “zoom” button to increase the granularity by which the range can be manipulated. If the user then wants to restore the control to the natural full range of values within the dataset, clicking the “reset” button will reset the LUT control to its original settings.
The user can change the color theme of the LUT by right clicking the mouse anywhere in the LUT bitmap. A menu of color themes will then appear from which the user can then change the display colors. The change will apply instantly to the bound display image (primary images are controlled by the primary LUT, and secondary images by the secondary LUT). In addition to manipulating and changing the LUT, the user can also invert the LUT and force it into a “reflective” mode, where it projects a mirror image along the top of the base LUT.
When the layout is anything but the MIP layout, moving the mouse along the image surface with the left mouse button pressed down will move a cursor through image space, auto-triangulating the other layouts (whether currently displayed or not). Moving the mouse with the right button pressed will scroll the current display windows (and may also open a context menu, depending on the capabilities of the underlying data and the program mode at that time).
By default, the program will automatically calculate an appropriate image zoom to fill the application window with the selected image in the selected layout. As the user resizes the application window, the program recalculates the display zoom and redraws the images. If the user wants to manually manipulate the level of zoom, a control for this purpose is embedded on the right side of the application windows (Fig. 2). They must first deactivate “auto-zooming” by de-selecting the “auto” zoom checkbox on the bottom of the control. The user can then manually manipulate the zoom level from 1 to 1000% of the image’s native size (in pixels).
Software Implementation: Image Quantification
The results of VOI sampling can be reviewed in one of two ways within the program, either through a tabular display where all VOIs and their corresponding statistical measures are presented in a spreadsheet-like table, or, in the case of dynamic data, as a real-time graph of single-measure results vs time in a time-activity curve (TAC). TACs can be exported into a variety of image formats (e.g., jpeg, tiff, and bmp). The user can enable/disable deferred image sampling in dynamic datasets by opening the context menu for the VOI (right-clicking within the VOI) and selecting/de-selecting the “smart refresh” option.
In addition to reviewing the quantitative results in real time, the Imager-4D has the ability to export results into a file for post-processing in a spreadsheet or other analysis system. Currently, the program allows the user to export VOI data as a tab-delimited text file which is easily importable into standard spreadsheet and statistical analysis software packages. In addition, the program allows for exporting the voxel-by-voxel VOI data on an individual VOI basis. In this case, a tab-delimited file is created and the voxel data is presented in a tabular structure on a row-by-row and slice-by-slice format, with empty cells representing voxels excluded by the VOI structure (as in the case of spheres).
Case Example of Software Utilization
As an example of the Imager-4D functionality, we have selected dynamic WB 2-deoxy-2-[18F]fluoro-d-glucose (18F-FDG) image data from a 50-year-old woman with metastatic colorectal cancer and thyroiditis. Spherical VOIs were placed within the normal parenchyma of the right lobe of the liver (liver-REF in subsequent text and figures), a colorectal cancer metastasis in the dome of the liver (liver-dome), a colorectal cancer metastasis in the inferior and posterior portion of the right lobe of the liver (liver-LB), the left lobe of the inflamed thyroid (thyroiditis-Lt), the right lobe of the inflamed thyroid (thyroiditis-Rt), and the blood pool of the left ventricle of the heart (cardiac). The VOIs were utilized to derive a number of radiomic and dynamic parameters from the WB 18F-FDG PET acquisition.
Figure 3 demonstrates the lean-body-mass-corrected maximum SUV for each of the VOIs as a function of time (liver-REF, black line with red circles; liver-dome, dark blue line with orange circles; liver-LB, red line with green circles; thyroiditis-Lt, green line with light blue circles; thyroiditis-Rt, blue line with blue circles; and cardiac, pink line with purple circles). Generating curves for each of a number of parameters involves selecting the desired parameter from the graphing-variable drop-down menu in the lower left corner of the results viewer window (red arrowhead in Fig. 3).
Fig. 3.
Analysis for a 50-year-old woman with metastatic rectal cancer and thyroiditis. Graphical display of maximum lean-body-mass-corrected SUV vs time for all selected VOIs (the color coding of the VOIs is listed in the text and at the bottom of the figure). The graphing-variable drop-down menu is delineated by the red arrowhead
Utilizing the same VOIs but selecting radiomic features from the graphing-variable drop-down menu shows the selected radiomic parameter as a function of time. A selected example is shown in Fig. 4 (in this case, IDM heterogeneity). For this example, the radiomic parameters were calculated using dynamic gray-level binning determined from a lean-body-mass-corrected SUV step function with each step equal to 0.2 SUV units and with a theta of 1 (i.e., textural features were derived considering only adjacent voxels). Corresponding radiomic parameters were also calculated for the underlying co-acquired CT image utilizing the same VOIs but using a step-function with a fixed 32 gray-level bins (not shown).
Fig. 4.
Graphical analysis of a selected radiomic feature. Graphical display of IDM homogeneity vs time for all selected VOIs (the color coding of the VOIs is listed in the text and at the bottom of the figure)
When utilizing the Patlak modeling feature of the Imager-4D, novel images can be created using correlation filtering. As noted previously, Patlak modeling produces a voxel-by-voxel goodness-of-fit parameter (Pearson correlation coefficient, “r”). Figure 5 a is a TCS configuration displaying positive correlation coefficients from the PET acquisition. Figure 5b is a TCS PET/CT configuration in which the lean-body-mass-corrected SUV display is filtered to only show those voxels that exhibited an r2 ≥ 0.85 in the Patlak fitting.
Fig. 5.
Patlak modeling image display. a TCS configuration displaying positive correlation coefficients (Pearson correlation coefficients, “r”) from the PET acquisition. b TCS PET/CT configuration in which the lean-body-mass-corrected SUV display is filtered to only show those voxels that exhibited an r2 ≥ 0.85 in the Patlak fitting
Discussion
The software described in this manuscript, Imager-4D, represents a potentially widely available means to speed the adoption of advanced metrics in PET imaging as routine adjuncts for patient risk-stratification and prognostication in the clinic. Although a great deal of complexity underlies the architecture of the software, the front-end user interface can be quickly learned and utilizes intuitive controls. As the Imager-4D is available for free download (https://jeffreyleal.wixsite.com/jleal) to interested clinician scientists, along with the de-identified example case in the previous section, this software may remove one of the barriers to widespread clinical adoption of dynamic and radiomic PET data in that it obviates the need for expensive or specialized stand-alone workstations to carry out the acquisition of such parameters.
We anticipate that the Imager-4D will evolve over time. Particularly with the recent attempt by the International Biomarkers Standardization Initiative to provide standardized image features and calculation methods in radiomics [17], and as the methodology for radiomics continues to improve, the code for the Imager-4D can be progressively updated. The same principle applies to dynamic WB imaging as new technology, such as the EXPLORER scanner [18], and improved reconstruction methods [19, 20] become accessible.
In the near-term future, the Imager-4D may provide a readily available means of combining information from both dynamic PET acquisitions and radiomics. Given the relatively recent advent of both of these methodologies, efforts at combining the two have not yet been reported. With the Imager-4D, it is not only possible to derive both dynamic and radiomic parameters from a PET study, but users can also obtain radiomic data from images such as the Patlak model positive correlation coefficient image in Fig. 5a. Such “dynomic” parameters have rarely been studied before [21], although the Imager-4D provides facile access to those factors and potentially opens up entirely new avenues of quantitative image analysis.
Most of the features of the Imager-4D are not specific to a particular radiotracer. Thus, new radiotracers that have either recently been U.S. Food and Drug Administration–approved and have entered clinical use (e.g., 18F-FACBC [22] or 68Ga-DOTATATE [23]), or investigational agents such as prostate-specific membrane antigen-targeted radiotracers [24], can also be studied with this software.
Finally, while the software has been developed and is readily available at our institution, much work remains to be done to aggregate a significant volume of WB dynamic PET data and to make use of the Imager-4D in routine data analysis. Although multiple clinicians and scientists at our institution have obtained useful data from the software, we have not conducted a formal usability analysis of the Imager-4D, a limitation of the current work. In due course, we will report on user experience and integration of this software into emerging research endeavors. Further, the Imager-4D is, at the current time, a research-only tool that lacks direct communication with either a picture archiving and communications system (PACS) or electronic medical record system, requiring its own back-end image database for DICOM archive and communication purposes. At a future date, incorporation of this software into routine clinical work for detailed image analysis and patient prognostication could be considered, although there would be regulatory hurdles to any such endeavor.
Conclusions
The Imager-4D is a novel PET viewer that allows the user to view dynamic PET data and to derive dynamic and radiomic parameters from that data. This includes the ability to combine dynamic data with radiomics (“dynomics”), potentially opening new avenues of PET quantitative analysis. The Imager-4D is available as a free download to interested clinician scientists, can obviate the need for stand-alone or specialized workstations, and may speed the adoption of advanced analysis of dynamic PET data into routine clinical use.
Funding Information
We gratefully acknowledge funding from the National Institutes of Health EB024495 and the National Cancer Institute P30CA006973.
Compliance with Ethical Standards
Conflict of Interest
The authors declare that they have no conflicts of interest.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Contributor Information
Steven P. Rowe, Email: srowe8@jhmi.edu
Jeffrey P. Leal, Email: jleal1@jhmi.edu
References
- 1.Karakatsanis NA, Lodge MA, Tahari AK, Zhou Y, Wahl RL, Rahmim A. Dynamic whole-body PET parametric imaging: I. Concept, acquisition protocol optimization and clinical application. Phys Med Biol. 2013;58:7391–7418. doi: 10.1088/0031-9155/58/20/7391. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Rahmim A, Lodge MA, Karakatsanis NA, Panin VY, Zhou Y, McMillan A, Cho S, Zaidi H, Casey ME, Wahl RL. Dynamic whole-body PET imaging: principles, potentials and applications. Eur J Nucl Med Mol Imaging. 2019;46:501–518. doi: 10.1007/s00259-018-4153-6. [DOI] [PubMed] [Google Scholar]
- 3.Cherry SR, Jones T, Karp JS, Qi J, Moses WW, Badawi RD. Total-body PET: maximizing sensitivity to create new opportunities for clinical research and patient care. J Nucl Med. 2018;59:3–12. doi: 10.2967/jnumed.116.184028. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Lee JW, Lee SM. Radiomics in oncological PET/CT: clinical applications. Nucl Med Mol Imaging. 2018;52:170–189. doi: 10.1007/s13139-017-0500-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Papp Laszlo, Rausch Ivo, Grahovac Marko, Hacker Marcus, Beyer Thomas. Optimized Feature Extraction for Radiomics Analysis of 18F-FDG PET Imaging. Journal of Nuclear Medicine. 2018;60(6):864–872. doi: 10.2967/jnumed.118.217612. [DOI] [PubMed] [Google Scholar]
- 6.van Helden EJ, Vacher YJL, van Wieringen WN, van Felden FHP, Verheul HMW, Hoekstra OS, et al. Radiomics analysis of pre-treatment [18F]FDG PET/CT for patients with metastatic colorectal cancer undergoing palliative systemic treatment. Eur J Nucl Med Mol Imaging. 2018;45:2307–2317. doi: 10.1007/s00259-018-4100-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Li K, Sun H, Lu Z, Xin J, Zhang L, Guo Y, Guo Q. Value of [18F]FDG PET radiomic features and VEGF expression in predicting pelvic lymphatic metastasis and their potential relationship in early-stage cervical squamous cell carcinoma. Eur J Radiol. 2018;106:160–166. doi: 10.1016/j.ejrad.2018.07.024. [DOI] [PubMed] [Google Scholar]
- 8.Fahrni G, Karakatsanis NA, Di Domenicantonio G, Garibotto V, Zaidi H: Does whole-body Patlak 18F-FDG PET imaging improve lesion detectability in clinical oncology? Eur Radiol, 2019 [DOI] [PubMed]
- 9.Constanzo J, Wei L, Tseng HH, El Naqa I. Radiomics in precision medicine for lung cancer. Transl Lung Cancer Res. 2017;6(6):635–647. doi: 10.21037/tlcr.2017.09.07. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Hatt M, Tixier F, Pierce L, Kinahan PE, Le Rest CC, Visvikis D. Characterization of PET/CT images using texture analysis: the past, the present… any future? Eur J Nucl Med Mol Imaging. 2016;44(1):151–165. doi: 10.1007/s00259-016-3427-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Im KC, Choi IS, Ryu JS, Eo GS, Kim JS, Moon DH. PET/CT fusion viewing software for use with picture archiving and communication systems. J Digit Imaging. 2010;23:732–743. doi: 10.1007/s10278-009-9227-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Leal J, Turkbey E, Solnes L, Rowe S, Rahmim A, Lodge M. A viewer for dynamic whole body PET/CT studies. J Nucl Med. 2017;58(Suppl 1):705. [Google Scholar]
- 13.Patlak CS, Blasberg RG, Fenstermacher JD. Graphical evaluation of blood-to-brain transfer constants from multiple-time uptake data. J Cereb Blood Flow Metab. 1983;3:1–7. doi: 10.1038/jcbfm.1983.1. [DOI] [PubMed] [Google Scholar]
- 14.Patlak CS, Blasberg RG. Graphical evaluation of blood-to-brain transfer constants from multiple-time uptake data. Generalizations. J Cereb Blood Flow Metab. 1985;5:584–590. doi: 10.1038/jcbfm.1985.87. [DOI] [PubMed] [Google Scholar]
- 15.Bentourkia M, Zaidi H. Tracer kinetic modeling in PET. PET Clin. 2007;2:267–277. doi: 10.1016/j.cpet.2007.08.003. [DOI] [PubMed] [Google Scholar]
- 16.Karakatsanis NA, Lodge MA, Zhou Y, Wahl RL, Rahmim A. Dynamic whole-body PET parametric imaging: II. Task-oriented statistical estimation. Phys Med Biol. 2013;58:7419–7445. doi: 10.1088/0031-9155/58/20/7419. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Hatt M, Vallieres M, Visvikis D, Zwanenburg A. IBSI: an international community radiomics standardization initiative. J Nucl Med. 2018;59(Suppl 1):287. [Google Scholar]
- 18.Cherry SR, Badawi RD, Karp JS, Moses WW, Price P, Jones T. Total-body imaging: transforming the role of positron emission tomography. Sci Transl Med. 2017;9:eaaf6169. doi: 10.1126/scitranslmed.aaf6169. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Zhang X, Zhou J, Cherry SR, Badawi RD, Qi J. Quantitative image reconstruction for total-body PET imaging using the 2-meter long EXPLORER scanner. Phy Med Biol. 2017;62:2465–2485. doi: 10.1088/1361-6560/aa5e46. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Karakatsanis NA, Casey ME, Lodge MA, Rahmim A, Zaidi H. Whole-body direct 4D parametric PET imaging employing nested generalized Patlak expectation-maximization reconstruction. Phys Med Biol. 2016;61:5456–5485. doi: 10.1088/0031-9155/61/15/5456. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Zhuang M, Karakatsanis NA, Dierckx RAJO, Zaidi H. Quantitative analysis of heterogeneous [18F]FDG static (SUV) vs. Patlak (Ki) whole-body PET imaging using different segmentation methods: a simulation study. Mol Imaging Biol. 2019;21:317–327. doi: 10.1007/s11307-018-1241-8. [DOI] [PubMed] [Google Scholar]
- 22.Savir-Baruch B, Zanoni L, Schuster DM. Imaging of prostate cancer using fluciclovine. Urol Clin North Am. 2018;45:489–502. doi: 10.1016/j.ucl.2018.03.015. [DOI] [PubMed] [Google Scholar]
- 23.Sanli Y, Garg I, Kandathil A, Kendi T, Zanetti MJB, Kuyumcu S, Subramaniam RM. Neuroendocrine tumor diagnosis and management: 68Ga-DOTATATE PET/CT. AJR Am J Roentgenol. 2018;211:267–277. doi: 10.2214/AJR.18.19881. [DOI] [PubMed] [Google Scholar]
- 24.Rowe SP, Gorin MA, Allaf ME, Pienta KJ, Tran PT, Pomper MG, et al. PET imaging of prostate-specific membrane antigen in prostate cancer: current state of the art and future challenges. Prostate Cancer Prostatic Dis. 2016;19:223–230. doi: 10.1038/pcan.2016.13. [DOI] [PMC free article] [PubMed] [Google Scholar]