Data explosion: the challenge of multidetector-row CT

https://doi.org/10.1016/S0720-048X(00)00270-9Get rights and content

Abstract

The development of multi detector-row CT has brought many exciting advancements to clinical CT scanning. While multi detector-row CT offers unparalleled speed of acquisition, spatial resolution, and anatomic coverage, a challenge presented by these advantages is the substantial increase on the number of reconstructed cross-sections that are rapidly created and in need of analysis. This manuscript discusses currently available alternative visualization tecvhniques for the assessment of volumetric data acquired with multi detector-row CT. Although the current capabilities of 3-D workstations offer many possibilities for alternative analysis of MCDT data, substantial improvements both in automated processing, processing speed and user interface will be necessary to realize the vision of replacing the primary analysis of transverse reconstruction′s with alternative analyses. The direction that some of these future developments might take are discussed.

Introduction

Without doubt, the greatest challenge of multidetector-row CT is dealing with ‘data explosion’. For our carotid/intracranial CT angiograms, we routinely have 375 images to review (300 mm coverage reconstructed every 0.8 mm); for aortic studies we have 450–500 images (∼600 mm coverage reconstructed every 1.3 mm); and for a study of the lower extremity inflow and run-off, we may generate 900–1000 transverse reconstructions [1]. While we could reconstruct fewer images for these data, experience with single-detector row CT scanners indicates that longitudinal resolution and disease detection is improved when at least 50% overlap of cross-sections is generated [2]. If we are to optimize our clinical protocols and take full advantage of these CT scanners, we will need to change the way that we interpret, transfer, and store CT data. Film is no longer a viable option. Workstation based review of transverse reconstructions for interpretation is a necessity, but the workstations must improve to provide efficient access to these data, and we must have a way of providing our clinicians with images that can be transported to clinics and the operating room. Alternative visualization and analysis using volumetric tools, including 3D visualization must evolve from luxury to necessity. We cannot rest on historical precedent to interpret these near isotropically sampled volumetric data using transverse reconstructions alone [3]. Although the tools for volumetric analysis on 3D workstations have evolved over recent years, they have probably not yet evolved to a level that routine interpretation can be performed as efficiently and accurately as transverse section review.

Both hardware and software developments must occur. While current computer workstations and visualization software are certainly adequate for assessing these MDCT data volumetrically, the process is very time consuming. What follows are a description of current workstation capabilities and a brief discussion of where development needs to go to facilitate the complete integration of volumetric analysis into the interpretive process of CT data.

Four main visualization techniques are currently in use on clinical 3D workstations — multiplanar reformation (MPR), maximum intensity projections (MIP), shaded surface displays (SSD), and volume rendering (VR). The first two techniques are limited to external visualization, while the latter allow for immersive or internal visualization and can be used for endoscopic-type applications.

Section snippets

Multiplanar reformation

MPR is a very convenient and available technique for displaying the data. One substantial limitation of traditional MPR is that visualized structures must lie in a plane. Because almost all structures for which 3D visualization is desired do not lie within a single plane, an MPR cannot be created that demonstrates the entirety of a structure. As structures course in and out of the MPR, pseudostenoses are created. The solution to this problem is to use curved planar reformations (CPR). Similar

Maximum intensity projection

MIPs are created when a specific projection is selected (anteroposterior for example) and then rays are cast perpendicular to the view through the volume data, with the maximum value encountered by each ray encoded on a 2D output image [4], [5]. As a result the entire volume is ‘collapsed’ with only the brightest structures being visible (Fig. 1 C). Variations of this approach include the minimum intensity projection (MinIP), which can be useful for visualizing airways [6], and the raysum or

Shaded surface display

Shaded surface displays provide exquisite three-dimensional representations of anatomy, relying on gray-scale to encode surface reflections from an imaginary source of illumination [9], [10] (Fig. 1D). The majority of SSDs created on clinical workstations display a single surface that is the interface between user-selected thresholds. As a result, the 12 bit CT data is reduced to binary data, with each pixel being either within or outside of the threshold range. Some workstations allow several

Volume rendering

The final and most complex rendering technique is VR [12], [13], [14], [15], [16] (Fig. 1E). There are many different versions and interfaces for VR, but the general approach is that all voxel values are assigned an opacity level that varies from total transparency to total opacity. This opacity function can be applied to the histogram of voxel values as a whole or to regions of the histogram that are classified as specific tissue types. With the latter approach, rectangular or trapezoidal

Editing

The challenge of performing efficient and accurate 3D visualization of clinical CT data is to balance the use of visualization techniques that require editing versus those that do not. In general it is preferable to avoid time-consuming editing, however this is unavoidable on the majority of clinical workstations in 2000. Editing can span from very quick and simple interactive cut-plane selection to meticulous 2D ROI selection with intermediate steps being provided by 3D ROI editing, region

Endoluminal visualization

The ability of helical CT to image the inner surfaces of tubular lumena has lead to proposed clinical applications of ‘virtual endoscopy’ to examine the bowel [16], [19], [20], airways [16], [21], [22], [23], blood vessels [16], [24], and urinary tract [25], [26]. Very little clinical validation of the utility of these techniques exist with the exception of colonography. While the term ‘virtual endoscopy’ is catchy, it is vague and loosely applies to any technique that displays the interior of

Room for improvement

The rapid advancement in computer processor speeds and the reduction in the cost of memory has resulted in relatively readily-available inexpensive CT workstations that are capable of performing all of the aforementioned visualization functions. However, they are still far from where they need to be to allow radiologists to abandon the standard approach of reviewing a stack of transverse CT reconstructions in favor of a truly volumetric exploration of the CT data. The advances need to come in

References (29)

  • S. Napel et al.

    STS-MIP: A new reconstruction technique for CT of the chest

    J. Comput. Assist. Tomogr.

    (1993)
  • H.E. Cline et al.

    3D surface rendered MR images of the brain and its vasculature

    J. Comput. Assist. Tomogr.

    (1991)
  • G.D. Rubin et al.

    Spiral CT of renal artery stenosis: comparison of three-dimensional rendering techniques

    Radiology

    (1994)
  • R.A. Drebin et al.

    Volume rendering

    Comput. Graphics

    (1988)
  • Cited by (0)

    View full text