Cinematic rendering – an alternative to volume rendering for 3D computed tomography imaging

Current computed tomography (CT) scanners provide isotropic submillimeter resolution enabling the utilization of various 2-D and 3-D post-processing techniques with excellent image quality. Whereas 2-D techniques such as planar or curved multiplanar reformations (MPR) as well as maximum intensity projection (MIP) images can be considered part of daily radiological routine, 3-D imaging post-processing techniques such as surface shaded display (SSD) and volume rendering (VR) are mainly used to convey complex anatomy. Here, 3-D visualization of CT image data can offer important information compared with 2-D axial images alone both for pre-operative planning and post-treatment follow-up.

While the generation of VR images in the early days was hampered by high computational demands and often was associated with long post-processing times with considerable user interaction, today’s powerful computers allow for a near-real time and automatic rendering of 3-D images in a fast, clinically compatible manner. Thus, VR is today used as the standard technique for the 3-D visualization of CT image data. The main advantages of VR lie in a relatively simple evaluation of different anatomical structures of a larger body region at the same time. Moreover, VR enables the display of CT image data in a coloured fashion [1, 2].

However, the VR technique also suffers from some disadvantages. While VR can be helpful in evaluating complex anatomy, it can also mask anatomical information and, hence, pathology. This has been recently illustrated for the visualization of intracranial aneurysms with VR, where 3-D CT angiography with bone subtraction provided a significantly higher diagnostic accuracy compared to 3-D CT angiography without bone subtraction [3]. Thus, in order to avoid missing critical information, VR images should always be correlated and compared with corresponding MPR images.

Finally, it should be kept in mind that visualization of imaging data with VR – as with all other post-processing techniques such as MIP or SSD—never adds information to the reader beyond what is present in the original source images. It always remains an option for a different representation of imaging data, which might be helpful for anatomically complex structures and disease, and for easier and comprehensible illustration of imaging findings to clinicians.

Recently, a new technique for 3-D visualization of cross-sectional image data from CT has been introduced, which is entitled cinematic rendering (CR). CR works with random sampling computational algorithms and uses different light maps to generate a realistic depiction of medical data. While first studies have started using the CR technique for visualizing anatomy from CT data [4, 5], there is so far—to our knowledge—no publication describing and discussing the new technique in more detail.

Thus, the purpose of this pictorial review is to illustrate the potential value of CR in comparison with conventional VR in various applications and body regions. For doing so, we (i) summarized the technical background and potential clinical applications of the conventional VR technique, (ii) introduced the technical concept of CR, and (iii) selected a number of common and some rare pathology and disease for visualizing the potential of the new 3-D post-processing technique for CT image post-processing.

The technical background of the VR technique has been repetitively explained in detail in a number of previous publications [2, 6]. VR represents a computer algorithm used to transform cross-sectional image datasets (from, e.g., CT or magnetic resonance (MR) imaging) into 3-D images. The VR technique consists of the following two steps: classification of each voxel and image projection. Classification determines how each point on the artificial rays that pass through the data contributes to the pixel value on the picture [7, 8]. For separation of different tissues (e.g., bone, soft tissue), a trapezoid is used for each tissue type. The tissue represented in each voxel of the volume dataset is determined by using predefined attenuation threshold levels and is assigned to a specific colour and opacity. Then, the weighted sum of the percentage of each tissue type represented in the voxel is calculated to determine the overall colour and transparency of each voxel. This step is performed for each voxel and for the whole volume dataset.

Then, the 3-D volume is displayed by using a projection technique. This is done by simulating rays of light, which are projected through the 3-D volume that contains the classified voxels. Each voxel, which is passed by the simulated rays, modulates the colour of the light depending on the assigned colour and transparency and contributes to the final projection and final image.

Finally, 3-D perception in the projected image can be enhanced by implementing additional effects, such as reflections and shadows on the surface of the rendered image. The final image can be manipulated by increasing or decreasing the slope of the trapezoid, which can be observed nearly in real time [8]. When sliding the trapezoid towards lower Hounsfield unit values, more structures with lower attenuation will be included (e.g., smaller vessels), and vice versa.

The development of CR was inspired by the entertainment industry based on the quality of computer animation programs in cinema, and had the aim to generate more photo-realistic representations of the human body from CT and MR image data sets.

CR is currently available as a research tool in an open research environment containing various prototype software (Frontier, version 1.0.0, syngo.Via, version VB10A, Siemens Healthcare, Forchheim, Germany) and has so far no approval for clinical use.

The CR technique [24] introduces a new paradigm, enabled by recent advances in computer graphics, to render volumetric medical image data by using a physically based real-time technique [25]. Conventional VR, such as ray casting [26] only considers emission and absorption of energy along a light ray to calculate 3-D images. Scattering effects are modelled using a local gradient shading model. Although simple to compute using a Riemann integral, such conventional approaches neglect complex light paths with multiple scattering patterns as well as light extinction, which might lead to a less artificial and potentially less accurate VR images.

As opposed to conventional VR methods, CR solves the multi-dimensional and non-continuous rendering equation [24] to integrate the light scattered from all possible directions along a ray. Thus, path tracing used in CT integrates a huge number of light rays, each with different paths to form each pixel of the rendered image. Since the number of light paths which can be traced is in theory infinite, and tracing of light paths is computationally expensive, Monte-Carlo simulations are used to generate a randomized subset of light paths with an adequate distribution. The final image is obtained iteratively by progressively averaging numerous Monte Carlo samples representing radiance at random positions with light scattered in random directions.

High dynamic range (HDR) rendering light maps are used for illumination, which lead to a natural illumination of the rendered data, in contrast to synthetic light sources used in conventional rendering methods such as VR [27].

As a result, the physically based VR method called CR computes in real-time the complex physics of lighting effects. It models shadows, ambient occlusion, multi-scattering, and color transmittance as well as sophisticated camera properties, which include concepts such as aperture, exposure and shutter speed. This approach leads to a natural and physically accurate presentation of the medical data, with a focus on an enhanced depth and shape perception [28].

The handling of transparency in CR does not differ from the conventional VR technique. The transparency is computed based on the transfer-function, which assigns colour and opacity to each attenuation value. Potential differences in transparency approaches can be easily adjusted by fine tuning of the transfer-function.

Representative image examples of CR in comparison with VR are provided in Figs. 1, 2, 3, and 4 for vascular anatomy and pathology, in Figs. 5, 6, and 7 for bones and skeletal disease, and in Figs. 8 and 9 for visualization of soft tissue anatomy. All CT source images were reconstructed with our default settings using a slice thickness of 2 mm (increment 1.5 mm), except for coronary CT angiography, which was reconstructed with a slice thickness of 0.75 mm (increment 0.5 mm). All CT angiography data were acquired after the intravenous injection of 50–80 ml non-ionic, iodinated contrast media (concentration 300–370 mg J/ml, depending on the body region).