Computed tomography (CT) was first introduced into radiologic practice in 1973, heralding the birth of a new (digital) era in diagnostic radiology. Credit for the development of this imaging technology is normally given to A.M. Cormack, a South African medical physicist working at Tufts University1 and G.N. Hounsfield, an engineer working at the Central Research Laboratories of the British company EMI.2 First generation CT scanners employed a synchronous translation of the x-ray source and sodium iodide detectors in 1° increments through a total of 180° around the patient's head. Data for two images could be acquired in about 4.5 minutes, with image reconstruction requiring an additional 20 minutes.3 The first CT scanners incorporated a water bath for the patient, and were only capable of scanning heads.

Despite the long scanning times and poor image quality, CT initiated a revolution in neuroimaging. As a recognition of the importance of CT in radiology, Cormack and Hounsfield shared the 1979 Nobel Prize for Medicine. By the 1980s CT had expanded into body imaging as well, and contrast agents were introduced to facilitate the visualization of vasculature, cerebrospinal fluid space, and the gastrointestinal tract. CT now plays an important role in the evaluation of many diseases. It has had a particular impact in the assessment of trauma, the detection of acute inflammatory processes, and the diagnosis, staging, and follow-up of neoplasms. CT is also used to guide a variety of diagnostic and therapeutic interventions.

First- and second-generation scanners demonstrated the clinical utility of CT scanning, but were very slow and cumbersome to use. During the late 1970s and early 1980s, third- and fourth-generation CT scanners were developed. Scanning times were reduced and image quality improved significantly. The introduction of slip ring technology in the 1980s paved the way for spiral CT scanning in the 1990s. The recent introduction of multi-slice CT will enable images to be obtained of a complete patient region in a matter of seconds. The relationship between the radiation dose imparted to the patient and the quality of the resulting image is complicated by the fact that the appearance of CT images has been divorced from the acquisition technique factors by the use of digital x-ray detectors and digital display devices. Understanding how CT image quality is related to acquisition technique factors should ensure that diagnostic information about the patient is maximized while reducing the corresponding radiation dose.

Computed tomography

Conventional CT4,5-CT images show slices (tomographs) through the patient. A typical CT slice has a 512 ¥ 512 matrix, with each pixel having a linear dimension of between 0.5 and 0.7 mm. The slice thickness defines the third dimension of the voxel (volume element), and ranges between 1 and 10 mm. The intensity value of each voxel, conventionally expressed in terms of Hounsfield Units (HU), represents the ability of the volume to absorb x-rays relative to water. By definition, water has a HU value of 0. Low density materials are less absorbing than water, and have negative HU values including air (-1,000 HU) and fat (-100 HU). High density and high atomic number materials are more absorbing than water, and have positive HU values such as bone (+1,000 HU).

A collimated fan x-ray beam passes through a patient and is recorded by an array of detectors, with the acquired data set called a projection. In modern CT scanners, one projection will typically have 700 to 900 discrete data points. For each acquired CT slice, approximately 1,000 projections are acquired as the x-ray tube rotates 360° around the patient. Image reconstruction is performed using a filtered back projection algorithm. The filter used in the reconstruction can be categorized as either a "detail" or "soft tissue" filter. Selecting the "detail" filter will improve the visibility of small image structures (better resolution), whereas the "soft tissue" filter will result in a smoother image by reducing the amount of visible mottle. The choice of reconstruction filter is important since it determines whether the reconstructed image emphasizes the visibility of small details or minimizes the amount of image mottle, thereby improving the visibility of low contrast objects (see Image Quality section for more detail).

The reconstructed image is stored as a two-dimensional array of numbers in a computer. Because CT is a digital imaging modality, the appearance of any voxel on a monitor or film is determined by the window/level setting selected by the operator. The choice of a window width and level creates a pair of thresholds, allowing the display to map pixel values less than the lower threshold to black, greater than the higher threshold to white, and between thresholds linearly to shades of gray. As an example, water with a HU value of 0 can take on the appearance of black, white, or any desired gray value in between. CT images may be printed onto film, viewed on a monitor display, or processed at a diagnostic workstation. In addition, the pixel values in a region of interest can be analyzed to obtain additional information about their distribution and their mean value.

Spiral CT6-In conventional CT, the patient lies on a table that remains stationary during the acquisition of projection data. The x-ray tube rotates 360° to acquire the projection data required to reconstruct the CT image at this patient location. During a spiral CT scan, however, the table moves during the rotation of the x-ray tube. The pitch ratio (PR) is the ratio of the table displacement during the x-ray tube rotation through 360° to the slice thickness. Spiral CT typically uses PR values between 1.0 and 2.0, although values < 1.0 and > 2.0 can be selected for specific clinical applications. The PR value selected for any given patient examination is very important, since it has a major effect on both image quality and patient radiation dose.

In spiral CT, the patient moves through the gantry as the x-ray tube performs a 360° rotation. Because of this patient motion, a complete set of projection data is not available directly for any axial slice through the patient. The necessary projections at any patient location are obtained by interpolation between the projection data acquired "upstream" and "downstream" at each angular location of the x-ray tube. Current spiral CT algorithms use a linear interpolation (LI) scheme, although non-linear schemes have also been investigated. In the early days of spiral CT, data was acquired up to 360° from the required slice location, but currently most interpolation algorithms use data acquired up to 180° from the required slice location.

The greatest benefit of spiral CT is the speed with which the patient image data are acquired. Spiral CT permits the chest or abdomen to be imaged during a single breath-hold, which results in fewer organ misregistration artifacts. In addition, spiral CT has the ability to track and image contrast agents, which permits arterial/venous phase imaging using one injection of iodine contrast. Another important benefit of spiral CT is increased flexibility when images are reconstructed. For example, the starting position of image reconstruction may be changed to any value and a judicious choice of the value of the image index can help to minimize problems of partial volume artifacts. Spiral CT can also help overcome the problem of "banding" or "stair-step" appearance of three-dimensional displays obtained from conventional axial images.

Image quality

For any medical imaging modality, image quality can be analyzed in terms of three distinct, but inter-related, components: contrast; spatial resolution; and mottle. A fourth component, temporal resolution, is also important in helical CT. In the sections that follow, contrast, spatial resolution, and mottle are explained for both conventional and spiral CT imaging.

Contrast-CT is a digital imaging modality, and has excellent low contrast resolution. For example, CT can readily differentiate between gray and white matter in the brain, which only differ in x-ray attenuation by about 0.5% (i.e., 5 HU). This performance is achieved by modifying the display contrast of CT images by adjusting the display window and level settings. As a result, lesion detection in CT images should never be limited because of inadequate image contrast. Conventional and spiral CT are capable of providing sufficient display contrast to permit most low contrast lesions to be observed. High mottle on CT images, however, can obscure the visibility of the lowest contrast lesions.

Spatial resolution-Spatial resolution is the ability to observe small details in images. Two distinct lesions, each 1 mm in diameter and separated by a distance of 1 mm, will appear as distinct points for an imaging system with good resolution. However, if the imaging system has a spatial resolution less than ~1 mm, these two objects will blur into one composite image. Spatial resolution is measured by a spatial frequency, expressed in line pairs per mm, where a line pair corresponds to one line that is "white" (or totally absorbing of x-rays) and one line that is "black" (or totally transparent to x-rays). If the lines have a thickness of 0.5 mm, the spatial frequency is 1 line pair per mm. The higher the spatial frequency that can be resolved, the better the imaging system in terms of spatial resolution performance.

Spatial resolution performance is important for high contrast objects, such as the bony structures of the inner ear where the intrinsic contrast between bone (+1000 HU) and air

(-1000 HU) is very high. CT spatial resolution performance is about 0.7 line pairs per mm, which is more than an order of magnitude worse than that of planar radiography. To maximize spatial resolution performance in CT, detail filters should be used in the reconstruction algorithms. Spatial resolution may also be increased by reducing the displayed field of view (i.e., by the use of a zoom feature), which can improve resolution by about a factor of two. The ultimate spatial resolution performance of a CT scanner is determined by the size of the focal spot and the size of the x-ray detectors. The in-plane (i.e., x-y plane) resolution of conventional and spiral CT are generally very similar.

In CT, the spatial resolution perpendicular to the imaged plane (z-axis) is determined by the slice thickness. The z-axis spatial resolution is much worse than the "in-plane" resolution (x-y plane), with typical values of 0.4 and 0.04 line pairs per mm for 1 and 10 mm slice thicknesses, respectively. The slice sensitivity profile (SSP) is intensity generated along the z-axis when a thin disc is imaged, and is used to characterize the amount of blur along the z-axis. Figure 1 shows the SSP for an axial CT scan as well as spiral CT scans acquired at pitch ratios of 1:1 and 2:1. Quantification of the broadening of the SSP is normally achieved by measuring the full-width half maximum (FWHM) distance, and/or the full-width tenth maximum (FWTM) distance. The greater the FWHM and FWTM values, the poorer the spatial resolution, and a corresponding loss in the ability of the imaging system to accurately reproduce fine details in the patient. Table 1 lists the FWHM and FWTM values for both conventional and spiral CT. High PR values increase the amount of blur introduced along the long patient axis, and reduce spatial resolution in this direction.

Mottle-A CT image of a uniform cylinder of water obtained with an ideal imaging system (i.e., no mottle present) would have every pixel value in the reconstructed image equal to zero, the HU value of water. For a real CT scanner, however, it is only the average pixel value in the image that is equal to zero. The water phantom image would have a mottled appearance, showing random fluctuation in individual pixel values about this nominal mean value. On a modern CT scanner, the water phantom image will have random fluctuation of ±3 HU about a mean value of zero. A total of 68% of all the pixels in the image will have intensity values between -3 and +3, 95% of pixels will have intensity values between -6 and +6, and 99% of pixels will have intensity values between -9 and +9.

To visualize two adjacent structures, their absolute difference in HU value must be greater than the mottle level. In other words, CT scanners can differentiate two tissues if their mean HU values are greater than the level of image mottle. Image mottle in CT is determined primarily by the number of x-rays used to generate a CT image. Table 2 shows how the mottle (i.e., standard deviation about the mean value) varies with the mean number of x-ray photons in a pixel. Image mottle can be reduced by increasing mAs, kV, and slice thickness, and can be minimized by the use of a soft tissue reconstruction filter. Mottle in spiral CT is reduced by ~20% when 360° LI is used, and is increased by ~12% when 180° LI is used. Differences in levels of image mottle between conventional and spiral CT scans are therefore relatively modest.

Patient doses

Radiation effects-The radiation received by a patient undergoing a CT examination is of interest since it permits an estimate to be made of (any) risk from the ionizing radiation. Risks from radiation can be classified as being either deterministic or stochastic effects.7 Deterministic effects include skin erythema and epilation, and are best quantified by the radiation dose to the specified organ. For deterministic effects there is a threshold dose below which the effect does not occur. Above the deterministic threshold dose, the severity of the radiation effect increases with increasing radiation dose. Stochastic effects include carcinogenesis and the induction of genetic effects in the offspring of irradiated individuals, and are quantified using the effective dose. For stochastic effects, it is generally assumed that there is no threshold dose below which there is no radiation related risk. The severity of stochastic effects is independent of radiation dose, but the probability of the effect occurring is taken to increase with increasing radiation dose.

Organ doses-It is of interest to know the dose to the skin, or to a specified organ, since this will quantify the possibility of inducing a deterministic effect. In CT, the skin directly in the acquisition field of view receives the highest radiation doses, with lower doses to tissues and organs on the central axis. Typical individual organ doses for head and body CT scans are summarized in Table 3. It is important to note that individual organ doses in CT are always below the threshold doses for the induction of deterministic effects. Even if patients undergo a series of CT examinations, eye cataracts and skin erythema cannot occur because the maximum organ doses are below the threshold dose of ~200 rem.

Patient risk (effective dose)8-In CT, as in most other types of radiological examinations that use ionizing radiation, the patient is subject to a complex three-dimensional dose distribution. The patient effective dose takes into account the mean dose to all irradiated organs and tissues, as well as their relative radiosensitivity. For head CT scans, patient effective doses are generally ~100 mrem, whereas for body CT scans, patient effective doses are generally ~500 mrem. Table 4 provides dosimetry data for other types of common radiological examinations that also use ionizing radiation.9 CT scans are the largest source of U.S. population radiation exposure from diagnostic radiology, because of high individual doses and a large number of CT scans performed. In the U.K., for example, CT scans account for ~2% of all diagnostic radiological examinations, but contribute 20% of the total dose to patients.10.11

Spiral CT dose-The dose to patients undergoing spiral CT scans depends on the selected values of PR and x-ray tube current (i.e., mA value). When the scanned patient range and tube current are the same, patient doses in spiral CT will be very similar to those of conventional CT, provided the PR is equal to 1.0. As the PR value is increased, however, the energy imparted to the patient will be reduced, which will lower the patient effective dose. For example, a PR of 2 for a single body examination would reduce the effective dose from ~500 mrem to ~250 mrem. The patient dose is also directly proportional to the selected mA value, and reducing the mA by 25%, will also reduce the patient effective dose by 25%.

In practice, there are several reasons why spiral CT will likely reduce patient doses. X-ray tube heat dissipation problems generally force operators to reduce x-ray tube currents, because the x-ray tube is being continually heated during a spiral CT examination. Fewer repeat CT scans will be required due to misregistration artifacts. Scans obtained to produce three-dimensional images will no longer require overlapping scans, since post processing of spiral CT data will generally be adequate for these types of procedure. Most important, however, is the fact that the patient dose is inversely proportional to the selected value of PR. In spiral CT, the selected PR can frequently taken on values of 1.5 or 2.0, which will ensure that patient doses are significantly lower than would be the case with conventional CT.

Conclusion

In spiral CT, contrast is essentially unchanged. The z-axis resolution in helical CT changes to an extent that is dependent on the selected pitch ratio, with higher PR values resulting in increased z-axis blur. Image mottle in spiral CT using current interpolation algorithms (i.e., 180° LI) is generally inferior to conventional CT, but these changes are relatively modest. Image reconstruction and image display in spiral CT are much more complex than in conventional CT. Spiral CT offers true three-dimensional information that can be processed, manipulated, and displayed to improve the ability of the radiologist to make the correct diagnosis. Spiral CT is also likely to result in lower radiation doses than conventional CT.

The optimum way of doing spiral CT scanning is task dependent.12,13 When an examination is to be performed, it is important to consider whether the lesion(s) to be detected are small or large, which will determine the relative importance of spatial resolution. It is also important to determine whether the lesion is low contrast or high contrast. For the former, it will be necessary to try to minimize image mottle, whereas for the latter, the presence of image mottle is unlikely to be an impediment to clinical diagnosis. In addition, the structured background, which could make lesion detection difficult, must be taken into account.

The major limitation in spiral CT scanning is the (limited) performance of current x-ray tubes. Modern x-ray tubes have large anode heat capacities and impressive heat dissipation rates, but they remain the weakest link in the CT diagnostic imaging chain. Future improvements in spiral CT will increase the x-ray utilization efficiency by the use of multi-slice detector technology. Current commercial multi-slice CT scanners permit up to four slices to be acquired simultaneously in a single rotation of the x-ray tube, effectively quadrupling the rate of data acquisition and the efficiency of x-ray utilization. In the near future, area detectors could permit the whole body region to be scanned in relatively short times. These are revolutionary advances for CT imaging, and could have the potential to replace many conventional radiographic examinations. Spiral CT is thus poised to have a major impact on medicine in the next decade. AR