AJR August 2017

U. Joseph Schoepf Corresponding Author Medical University of South Carolina Charleston, SC

“Cinematic Rendering in CT: A Novel, Lifelike 3D Visualization Technique”

Newly developed “cinematic rendering” technology can produce photorealistic 3D images from traditional CT and MRI data, with potential applications in medical education, communication with patients and physicians, and early disease detection. In this interview, Dr. U. Joseph Schoepf discusses the development and possible uses of this technology.

1. How would having 3D-rendered images change the way radiologists communicate with referring physicians and patients?

Radiologists are extensively trained in evaluating images in the classic format of grayscale-stacked slices in 2D, which are then developed into a 3D impression of the anatomy without a real need to look at 3D rendered images. However, for referring physicians, and particularly patients, understanding 3D anatomy and pathology from the 2D reconstructions is much more challenging, potentially resulting in miscommunications. In contrast, 3D rendered images make it easier to understand anatomy and pathology. All pertinent information is presented in the reconstruction and there is no need for further mental visualization. For instance, while it is easy for an experienced radiologist to follow bypass grafts on classic grayscale 2D reconstructions, referring physicians and patients would benefit from 3D volume rendered images as the entire anatomy is present and color scales simplify the differentiation of structures.

In that regard, we believe Cinematic Rendering (CR) provides two major advantages over classic volume rendering techniques. First, it has the ability to create very realistic images of the anatomy in terms of the 3D “feel” of the spatial relationship between structures. Second, through its variety of light maps, CR allows the user to highlight certain anatomical structures more clearly. In addition, through its transfer function, a wide variety of colors can be applied to further enhance the delineation of anatomical structures.

Of note, at the time of article drafting, the available software version did not allow for manual color determination. However, with the current version, it is now possible to manually edit the resulting curve of the transfer function, thus manually applying different colors to different structures. Together, the advantages of 3D CR make anatomy and pathological findings much more comprehensible to referring physicians and patients, effectively enhancing both physician-physician and physician-patient communication.

2. As this technology grows and develops, what will be some of the challenges faced by radiologists in making the most effective use of cinematic rendering?

As radiologists are trained to recognize disease on classical 2D grayscale reconstructions, 3D reconstructions are mostly seen as optional and are not often used in clinical workflows. While CR could eventually replace classic volume rendering as the main 3D reconstruction technique in settings where 3D reconstructions are clinically relevant, such as analysis of the coronary tree, the next objective in implementing CR is to identify new clinical scenarios that would benefit from its application.

For example, a scenario in which significant diagnostic relevance is provided by CR when added to classic 2D cross-section reconstructions, compared with classic volume rendering techniques alone. In that regard, further clinical studies are necessary to establish the diagnostic relevance of CR.

3. A limitation on this new technology is the computing power and time necessary to fully render these images. How can radiology departments overcome these and demonstrate that this technology is cost-effective?

3D reconstruction techniques require high computational power, as the anatomical information as a whole is included in the image. Moreover, CR necessitates more computational power than the classic volume rendering technique, as it simulates the pathways of billions of photons through anatomical volume. However, this can be improved in many ways.

If technological advancement trends continue at their current rate, workstations will continue to become more powerful at cheaper costs, allowing heavier software such as CR to run faster and more efficiently. Moreover, Graphics Processing Unit (GPU)-based processing is a relatively new concept that would allow the user to optimize CR’s performance. Indeed, a GPU compared to the Central Processing Unit (CPU) currently used for most computational operations contains a higher number of processing cores capable of performing many parallel computations and hence is more efficient for computationally intensive operations and physics simulations. Cinematic Rendering is a GPU-based application. And as graphics cards are constantly improving and used in an ever-increasing field (General Purpose Computation on Graphics Processing Unit [GPGPU]), we can expect that the trend continues constantly increasing hardware’s speed.

Moreover, CR’s code is constantly being optimized, both for the renderer core and of the Syngo via Frontier prototype. Thus, with future hardware generations that are not far away and algorithm optimization, real-time rendering is certainly conceivable. In the near future, a web-based, user-friendly CR application will be available, where radiologists, clinicians, instructors, and researchers will be able to upload datasets and perform 3D renderings with the CR algorithm. These renderings will be cloud- and GPU-based and thus will run faster and more efficiently.

Do you have a story idea for InPractice?
Contact Cheryl S. Merrill, Director of Publications, at cmerrill@arrs.org for more information.