Abstract
A methodology is described using Adobe Photoshop and Adobe Extendscript to process DICOM images with a Relative Attenuation-Dependent Image Overlay (RADIO) algorithm to visualize the full dynamic range of CT in one view, without requiring a change in window and level settings. The potential clinical uses for such an algorithm are described in a pictorial overview, including applications in emergency radiology, oncologic imaging, and nuclear medicine and molecular imaging.
Keywords: CT window settings, Grayscale visualization, Contrast sensitivity, CT dynamic range
Background
The dynamic range acquired in a typical computed tomography (CT) scan is wider than can be displayed on a conventional 8-bit or 10-bit grayscale monitor, or by many medical display applications. A 10-bit display system can theoretically display up to 1024 gradations of gray, and under ideal circumstances, the human eye can differentiate 720 just noticeable differences in gray shades with a calibrated clinical medical monitor that can display a luminance range between 0.8 and 600 cd/m2 [1]. A CT scan may have greater than 2000 gradations of gray quantified as Hounsfield units (HU) ranging in attenuation between −1000 HU (air) to >1000 HU (cortical bone). In order to visualize adequate contrast resolution on a workstation monitor and to maximize the contrast resolution of structures with similar attenuation values, it is necessary for the interpreting radiologist to review the CT data in multiple linear window settings, typically optimized for bone, lung, and soft tissue.
Through image processing, the dynamic range of a CT image can be compressed to a limited number of grayscale shades, maximizing the useful visual information that can be displayed on a single image, without the need to change window settings. Multiple methods can be used, including non-linear CT windows [2], histogram equalization [3], adaptive histogram equalization [4], and contrast-limited adaptive histogram equalization [5]. Because these algorithms operate upon the entire dynamic range acquired by CT, some grayscale ranges may be perceived with increased contrast in the processed result, while others may appear decreased. Several of these algorithms were developed in the relative infancy of CT and none have found widespread clinical use to date.
The concept of blending multiple CT window settings was first described in 2009 termed “Universal Trauma Window [6].” This developed further into Threshold Windowing (TW) that overlaid bone, lung, and soft tissue windows, and which comprised the single series sent from the combat environment (at the time an Air Force Theater Hospital in Balad, Iraq) highly compressed as an mp4 file [7].
One potential problem with these methods is that they may fundamentally change the standard appearance of the CT data in a way that may require adaptation by the radiologist. Fundamental relationships may not always be consistent or may become less conspicuous. For instance, radiologists have been trained that fat should always attenuate less (appear darker) than soft tissue, and lung should likewise attenuate less than fat; however, published examples of prior algorithms do not universally maintain these relationships [2–6].
The Relative Attenuation-Dependent Image Overlay (RADIO) window blending algorithm is described as a method to visualize the full dynamic range of CT without needing to change window settings. The RADIO window blending algorithm produces images that require very little retraining by the radiologist by maintaining the relative attenuation relationships between the fundamental anatomic densities, with progressively increasing grayscale values corresponding to air, fat, fluid, soft tissue, and bone. Up to three standard CT window settings, optimized for soft tissue, bone and lungs, can be blended to allow simultaneous visualization of these structures. This algorithm was conceptualized and implemented in readily available commercial software (Adobe Photoshop; Adobe Systems Incorporated, San Jose, USA). Through the use of a freely available (although closed source) Javascript-based scripting software (Extendscript), the algorithm can be applied to entire DICOM data sets, which can then be viewed using standard PACS workstation software.
Methods
The algorithm was created in Adobe Photoshop, which is a widely available commercial image editing software application that can read and write DICOM images, and has been previously discussed in the radiology literature [8]. Photoshop is most well known amongst radiologists for image manipulation [8–10], as well as sophisticated image processing such as automatic segmentation of MRI images [11]. The advantages of using Photoshop include wide availability across PC and Macintosh platforms, ability to perform rapid prototyping of algorithms, and ability to perform image processing without requiring specialized knowledge of computer programming. In conjunction with the freely available Javascript-based Adobe Extendscript utility, Photoshop can be scripted to process a full set of DICOM images.
The RADIO window blending algorithm is similar in concept to high dynamic range (HDR) photography using multiple exposure bracketing [12, 13], but modified for CT images. In HDR photography, multiple exposures (typically 3) of a scene are obtained at different shutter speeds, resulting in an underexposed image, a properly exposed image, and an overexposed image. The dynamic range of the underexposed image is optimized for highlights; the properly exposed image is optimized for the mid-range; and the overexposed image is optimized for the shadows. These three images are then combined in software using a technique called tone mapping to create a single HDR image with much greater dynamic range in comparison to a conventional photograph.
An analogous process can be modified for CT images, which forms the basis of the RADIO algorithm. The three “exposures” include CT window settings optimized for bone (“underexposed”), soft tissue (“properly exposed”), and lung (“overexposed”). These three windows are then blended in an attenuation-dependent manner, based on the relative gray scale values for each window setting [Fig. 1]. For instance, the lung window is blended to the soft tissue window only in areas of the image that are very low in attenuation (black); the higher attenuation regions of the lung window (which typically appear white or nearly white) are discarded. Likewise, the bone window is blended with the other windows only in areas that are relatively high in attenuation (white).
Fig. 1.
Diagram demonstrating the concept of the RADIO window blending algorithm, which blends the regions of an image that are not overlapping in attenuation
The algorithm imports each slice of the original DICOM data three times, once each for soft tissue, lung, and bone window settings, and a new Photoshop document is created with each window setting placed onto a unique layer. While Photoshop converts 10- or 12-bit DICOM images into a 16-bit format internally, the window settings must be set at the time of import, and Photoshop discards all image data that is outside of the dynamic range of the chosen import window and level settings. Therefore, the resultant Photoshop file contains less data than the original 10- or 12-bit DICOM file despite being in 16-bit format. This limitation in Photoshop’s handling of DICOM images does not affect image quality in the chosen window settings, but it does mean that the window settings cannot be reliably adjusted after import has occurred.
With the soft tissue window forming the primary layer, each of the layers is processed such that the “black” areas of the soft tissue window are replaced with the corresponding (low attenuating) areas of the lung window, and the “white” areas of the soft tissue window are replaced with the corresponding (high attenuating) areas of the bone window [Fig. 2]. This results in a combined image in which soft tissues are highly visible along with low and high attenuating tissues individually visualized best in lung and bone windows, respectively.
Fig. 2.
Diagram demonstrating the RADIO window blending algorithm. The three conventional CT window settings in soft tissue, bone, and lung windows are placed on separate layers in a Photoshop document. The soft tissue window layer is not modified. On the bone window layer, only relatively high attenuation information is retained, which replaces the corresponding pixel locations in the soft tissue window. On the lung window layer, only relatively low attenuation is maintained, which is summed with the underlying soft tissue window
The bone and lung windows are optimally obtained from the sharp CT reconstruction kernel for optimized bone/lung detail; if these are not available for a particular dataset, then the soft tissue window can be utilized. The means by which these layers are blended is achieved through Photoshop’s Advanced Blending options [Fig. 3], which controls the opacity of each pixel of an individual layer, dependent on its absolute post-window setting grayscale value. The absolute grayscale value of each pixel is dependent both on the inherent CT attenuation and CT window settings. The threshold value at which the blending is performed is a graduated range rather than an absolute cutoff, which allows for a smoother and more natural looking transition between the blended layers. In addition to the relative attenuation-dependent blending described above, the overall opacity of the lung and bone window layers is adjusted to minimize noise inherent in the sharp kernel window. A representative axial image of a normal CT of the chest [Fig. 4] demonstrates the RADIO window blending algorithm, with the component conventionally windowed CT images.
Fig. 3.
Screenshot from Adobe Photoshop demonstrating the advanced blending settings for the lung window layer (left image) and the bone window layer (right image)
Fig. 4.
a Axial contrast-enhanced CT of a normal chest demonstrating the RADIO window blended algorithm. Key features to note include relative good visualization of the osseous structures, the lungs, and the soft tissues. b Conventional soft tissue, c bone, and d lung window settings comprising the blended image
The algorithm just described will transform a single image. However, with the use of the Javascript-based Adobe Extendscript application, it is possible to automate this process and apply it to an entire DICOM data set. It is therefore possible to read and write DICOM images directly from a PACS server, and the modified images can be viewed on a standard PC with DICOM viewing software or the images can be sent back to the PACS depending on the configuration. The script requires the location of two folders—the folder containing the sharp kernel CT reconstructions (used as the bases for the lung and bone windows), and the folder containing the soft tissue kernel CT reconstructions—as input. The script then applies the algorithm to each image sequentially. In order to work properly, both folders must contain the exact same number of images. If only one kernel is available, then the same folder can be chosen for both. Using a freely available Scriptlistener plugin, which is included in the default installation of Adobe Photoshop, every Photoshop command is written to a log file in a format that can be scripted by the Extendscript environment. This log file-recorded information can then be inserted in the script.
On a standard laptop computer running MacOS (Apple Inc., Cupertino, CA) version 10.1.1 with a 2.3 GHz Core i7 (Intel Inc., Santa Clara, CA) quad-core central processor unit and 8 gigabytes of random access memory, image processing time is approximately 1.5 s per image, so a complete CT dataset can be processed in about 3 min. Since Photoshop runs equally well on Windows (Microsoft Inc., Seattle, WA), performance on equivalent Windows systems should be similar. Of note, Extendscript is an inherently inefficient method to perform image processing, as it is merely a way to tool to control Photoshop. Theoretically, it would be possible to dramatically improve the performance of the algorithm if dedicated image processing code were to be written.
Results
There are several potential clinical applications of the RADIO window blending algorithm. Examples in the fields of emergency radiology including combat settings as previously mentioned, oncologic imaging, and nuclear medicine and molecular imaging will be briefly discussed.
Emergency Radiology
In emergency radiology, it is necessary to rapidly and accurately interpret CT imaging of the acutely traumatized patient. Polytrauma patients frequently receive whole body CT for the evaluation of multiple organ injuries. At many tertiary trauma centers, the radiologist reviews the images in real time to identify all life threatening injuries and to immediately determine if additional images are necessary during the delayed or excretory phases. Critical injuries may present in brain (e.g., intracranial hemorrhage), lung (e.g., pneumothorax), soft tissue (e.g., mediastinal hematoma, vascular, or airway injury), or bone (e.g., fracture) CT window settings. Using current standard of care imaging, the CT dataset is typically reviewed in at least four windows to accurately report the spectrum of potential injuries. Basing critical clinical decision making on this “rapid first pass” interpretation of multiple injuries increases the potential for error due to the large number of images requiring review in multiple different window settings, especially in situations where the preliminary interpretation may be performed by a trainee.
As the RADIO algorithm is able to visualize the dynamic range of CT data with a single window setting, it may be possible to reduce the time necessary to provide an accurate preliminary interpretation in the acutely traumatized patient, especially in mass casualty [14] or combat settings [15]. In these critical situations, it is necessary to perform initial “dam-age control imaging” to direct patient triage contemporane-ously with patient scanning and initial management decisions. After initial patient stabilization, careful review of the CT can be performed using traditional window settings. Prior versions of the algorithm were used to assess ballistic trajectories of all organs (e.g., ribs, lung, vasculature) in a single pass during mass casualties. Figure 5 demonstrates a cervical spine frac-ture with an associated epidural hematoma. Figure 6 demonstrates a left anterior pneumothorax and a subtle posterior rib fracture. Figure 7 demonstrates a simple modification of the RADIO algorithm to utilize brain (rather than soft tissue) and bone windows to visualize an epidural hematoma and an overlying calvarial fracture.
Fig. 5.
a Axial noncontrast CT of the cervical spine at the level of C2 with the RADIO window blending algorithm applied, using soft tissue and bone source images. There are fractures of C2 involving the left pedicle and right facet (white arrows). There is a subtle epidural hematoma (black arrows point to the interface of the hematoma and the CSF), causing spinal stenosis. b Conventional soft tissue and c bone window settings comprising the blended image
Fig. 6.
a Axial contrast-enhanced CT of the chest in a trauma patient with RADIO window blending algorithm applied, using soft tissue, bone, and lung source images. There is a small left anterior pneumothorax (white arrow), and a subtle nondisplaced fracture of a left posterior rib (black arrow). b Conventional soft tissue, c bone, and d lung window settings comprising the blended image
Fig. 7.
a Axial noncontrast CT of the head with RADIO window blending applied, using brain and bone windows. There is a large right epidural hematoma (white arrows) and subtle overlying temporal bone fractures (gray arrows). b Conventional brain and c bone window settings comprising the blended image
Oncologic Imaging
For tumor staging and pre-operative planning prior to oncological surgery, it is necessary to understand the relationship of a tumor to the surrounding soft tissue and osseous structures. Often, tumors may involve structures that are of highly different attenuations (e.g., lung and bone, or bone and soft tissues), and creation of a single window to assess the tumor and its surrounding structures could optimize visualization. Since there is often debate as to which window to measure lung lesions (soft if adjacent to hila or pleura, lung if surrounded by lung), a single window may be the best solution.
Additionally, in multidisciplinary conferences, non radiologists and even patients themselves may find viewing of the blended images to be more intuitive to understand than viewing multiple windowed images of the same anatomic area. Figure 8 demonstrates a right hilar lung cancer with narrowing of the bronchus intermedius. Figure 9 demonstrates a left lower lobe mass.
Fig. 8.
a Axial contrast-enhanced CT of the chest in a patient with lung cancer with RADIO window blending algorithm applied, using soft tissue, bone, and lung source images. There are enlarged right hilar lymph nodes (white arrows), causing narrowing of the right bronchus intermedius (black arrow) and mild post-obstructive atelectasis (gray arrow). b Conventional soft tissue, c bone, and d lung window settings comprising the blended image
Fig. 9.
a Axial contrast-enhanced CT of the chest in a patient with a soft tissue sarcoma with RADIO window blending algorithm applied, using soft tissue, bone, and lung source images. There is a left lower lobe mass (arrow), which is closely associated with a pulmonary vessel. b Conventional soft tissue, c bone, and d lung window settings comprising the blended image
Nuclear Medicine and Molecular Imaging
Simultaneous acquisition of molecular data with positron emission tomography (PET) and anatomic data with CT allows precise localization of subcellular processes interrogated by the chosen radiopharmaceutical. In addition to performing primary interpretation of the PET, CT, and fused PET/CT images on a dedicated workstation, additional limited bit-depth images are typically created and stored on the PACS server for clinical reference. These fused images are comprised of CT data in a soft tissue window setting with superimposed color-mapped PET data. Although useful for general overview of the FDG-avid regions, such images cannot be windowed to the degree necessary for visualization of other structures such as lungs and bones. Application of the RADIO window blending algorithm to the CT data in order to create a fused image would allow more information to be displayed. Figure 10 demonstrates a fused PET-CT image in a patient with lung cancer and mediastinal metastases. Figure 11 demonstrates a fused PET-CT image in a patient with a pleural-based mass.
Fig. 10.
a Axial fused PET-CT image in a patient with lung cancer with RADIO window blending algorithm applied, using soft tissue, bone, lung, and PET source images. There is an FDG-avid right hilar mass and bilateral FDG-avid hilar lymph nodes. b Conventional soft tissue, c bone, and d lung window settings comprising the blended image
Fig. 11.
a Axial fused PET-CT image with RADIO window blending algorithm applied, using soft tissue, bone, lung, and PET source images. There is an FDG-avid pleural-based mass-like opacity in the superior segment of the right lower lobe (arrow), representing metastatic disease in a patient with squamous cell carcinoma of the mouth treated previously with radiation and chemotherapy. b Conventional soft tissue, c bone, and d lung window settings comprising the blended image
Discussion
The RADIO window blending algorithm maximizes the useful visual information displayed on a single image, allowing assessment of lung, soft tissue, and bone simultaneously. The natural relationships between the attenuation of these different tissues are maintained despite the compression of dynamic range. Therefore, it takes very little time for the reader to adapt to the algorithm. Because the radiologist does not need to change windows several times during the review of an examination, time can potentially be saved. Additionally, it may be possible to see relationships between anatomically adjacent structures that have wide differences in attenuation, such as between a mediastinal tumor and the adjacent lung, or a lung tumor and the adjacent bone. In additional to potentially reducing the time necessary to interpret a CT study, the RADIO blended images may provide an optimized way for non-radiologist physicians to visualize the relevant clinical data in one image, to display CT data for teaching or multidisciplinary conferences, or to directly show a patient their own CT data in a method that is highly intuitive.
One limitation of the RADIO window blending algorithm is the introduction of an “edge” or “halo” artifact at the interfaces of the blended lung and soft tissue windows. This is due to the fact that an object visualized in a soft tissue window will appear slightly larger than the same object visualized in a lung or bone window [16], due to the volume averaging of attenuation values at the periphery of the lesion. This phenomenon contributes to a “halo” appearance at the interface of lung and soft tissue windows, somewhat similar in appearance to (although completely unrelated in mechanism from) the India-ink artifact seen in chemical shift MR imaging when fat and water are present in the same voxel. The appearance of the “halo” artifact in RADIO window blended images is dependent on the source image. Most commonly, the artifact appears as a sharp demarcation at the pleural surface. When a soft tissue density object is surrounded by lung (e.g., a lung tumor), the soft tissue lesion may appear to have a “halo” surrounding it. One potential advantage of this artifact is to provide the radiologist a demarcation between window settings, which may potentially be useful to standardize measurement of lung tumors for clinical use or in cancer trials.
Due to limitations in Photoshop’s handling of imported DICOM data as previously discussed, the window and level settings must be set at the time of import, and therefore the image file produced is not able to be windowed to the same degree as raw CT data. However, since the window settings are already optimized, this may not be such a disadvantage in practical use. Previous research has shown that an optimized fixed window is equivalent to a freely adjustable window when evaluating the lungs [17]. Even if Photoshop could save the full bit depth of the imported DICOM data, the DICOM format [18] does not allow different layers of image data to comprise a single image. However, if the RADIO window blending algorithm were implemented into a PACS workstation by a vendor to circumvent these limitations, then each layer could retain the ability to be individually windowed.
Conclusion
We have described an algorithm to optimize display of the inherently high dynamic range acquired in a CT scan, in order to maximize the useful visual information displayed without needing to change window settings. This was achieved through Adobe Photoshop, a widely available image processing software, and Adobe Extendscript, a Javascript-based scripting language. The algorithm blends three individual window settings optimized for soft tissue, bone, and lung, and produces images that require very little adaptation by the radiologist. A simple modification of this basic algorithm can produce brain-bone blended images as well. Potential applications in emergency radiology, oncologic imaging, and nuclear medicine and molecular imaging were discussed. Potential future developments include clinical validation and vendor integration into CT consoles and PACS workstations.
Compliance with ethical standards
Disclosure
Les R. Folio is an associate investigator in a research agreement with Carestream Health (Rochester, NY).
References
- 1.Kimpe T, Tuytschaever T. Increasing the number of gray shades in medical display systems—How much is enough? J Digit Imaging. 2007;20(4):422–432. doi: 10.1007/s10278-006-1052-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Gomori JM, Steiner I. Non-linear CT windows. Computerized Radiology. 1987;11(1):21–27. doi: 10.1016/0730-4862(87)90025-4. [DOI] [PubMed] [Google Scholar]
- 3.Lehr JL, Capek P. Histogram equalization of CT images. Radiology. 1985;154(1):163–9. doi: 10.1148/radiology.154.1.3964935. [DOI] [PubMed] [Google Scholar]
- 4.Pizer SM, Zimmerman JB, Staab EV. Adaptive grey level assignment in CT scan display. Journal of Computer Assisted Tomography. 1984;8(2):300–5. [PubMed] [Google Scholar]
- 5.Fayad, L. M., Jin, Y., Laine, A. F., Berkmen, Y. M., Pearson, G. D., and Freedman, B.: Technical Developments Chest CT Window Settings with Multiscale Adaptive Histogram Equalization : Pilot Study. Radiology. 845–852,2002 [DOI] [PubMed]
- 6.Whitworth S, Folio L, Nguyen B.: Universal trauma CT window for rapid preliminary interpretation. Presented at the annual meeting of the American Society of Emergency Radiology 2009, referenced in: Emerg Radiol 16:502–503,2009
- 7.Peterson PG, Pak SK, Nguyen B, Jacobs G, Folio L. Extreme compression for extreme conditions: Pilot study to identify optimal compression of CT images using MPEG-4 video compression. Journal of Digital Imaging. 2012;25(6):764–770. doi: 10.1007/s10278-012-9500-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Caruso RD, Postel GC. Image editing with Adobe Photoshop 6.0. Radiographics. 2000;22(4):993–1002. doi: 10.1148/radiographics.22.4.g02jl32993. [DOI] [PubMed] [Google Scholar]
- 9.Caruso RD, Postel GC. Image annotation with adobe photoshop. Journal of Digital Imaging. 2002;15(4):197–202. doi: 10.1007/s10278-002-0017-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Corl FM, Garland MR, Lawler LP, Fishman EK. A Five-Step Approach to Digital Image Manipulation for the Radiologist. Radiographics. 2002;22(4):981–992. doi: 10.1148/radiographics.22.4.g02jl33981. [DOI] [PubMed] [Google Scholar]
- 11.Park JS, Chung MS, Hwang SB, Lee YS, Har DH. Technical report on semiautomatic segmentation using the Adobe Photoshop. Journal of Digital Imaging. 2005;18(4):333–343. doi: 10.1007/s10278-005-6704-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Jacobs A. High Dynamic Range Imaging and its Application in Building Research. Advances in Building Energy Research. 2007;1(1):177–202. doi: 10.1080/17512549.2007.9687274. [DOI] [Google Scholar]
- 13.Martínez MA, Valero EM, Hernández-Andrés J. Adaptive exposure estimation for high dynamic range imaging applied to natural scenes and daylight skies. Applied Optics. 2015;54(4):B241. doi: 10.1364/AO.54.00B241. [DOI] [PubMed] [Google Scholar]
- 14.Singh AK, Ditkofsky NG, York JD, Abujudeh HH, Avery LA, Brunner JF, et al. Blast Injuries: From Improvised Explosive Device Blasts to the Boston Marathon Bombing. RadioGraphics. 2016;36:295–307. doi: 10.1148/rg.2016150114. [DOI] [PubMed] [Google Scholar]
- 15.Folio L.: Combat Radiology, Diagnostic Imaging of Blast and Ballistic Injuries. Springer. 35–36,2010
- 16.Harris KM, Adams H, Lloyd DC, Harvey DJ. The effect on apparent size of simulated pulmonary nodules of using three standard CT window settings. Clinical Radiology. 1993;47(4):241–4. doi: 10.1016/S0009-9260(05)81130-4. [DOI] [PubMed] [Google Scholar]
- 17.Maguire WM, Herman PG, Khan A, Smith RH, Noma S, Eacobacci TM, Carlin M. Comparison of fixed and adjustable window width and level settings in the CT evaluation of diffuse lung disease. Journal of Computer Assisted Tomography. 1993;17(6):847–52. doi: 10.1097/00004728-199311000-00002. [DOI] [PubMed] [Google Scholar]
- 18.Larobina M, Murino L. Medical image file formats. J. Digit. Imaging. 2014;27:200–6. doi: 10.1007/s10278-013-9657-9. [DOI] [PMC free article] [PubMed] [Google Scholar]