Читать книгу Fractures in the Horse - Группа авторов - Страница 87

Image quality is a broadly understood concept, and assurance of diagnostic quality is critical to accurate use of medical imaging. There are several measures of image quality that are common to all techniques. Understanding these and their interactions aids in the recognition of a high‐quality image and variation from this.

Contrast is the greyscale value difference between adjacent regions on the image. On the final image, this is determined by a number of factors including the inherent subject contrast, detector contrast and displayed contrast. Subject contrast is determined by the tissues and the type of energy (radiation, sound wave and signal intensity) recorded. Detector contrast refers to the way that an input signal is converted to an output or a recorded signal. In most digital radiography (DR) systems, the characteristic curve or relationship between the energy of the X‐rays hitting the detector and the emitted or recorded image is close to linear. A linear characteristic curve without image processing would appear very ‘flat’ or washed out. Almost all digital imaging systems therefore process the output so that contrast is increased and the displayed image has a non‐linear output. Displayed contrast simply refers to the ability of the end user to manipulate the greyscale so that the image can have more or less contrast as desired. To put this into context, in order to identify a fracture on radiographs, subject contrast would be the variation in tissue density between the fracture gap and the fracture margins, the detector contrast would be determined by the settings of the radiographic system and its processing, and the displayed contrast would be chosen by the observer at the viewing station. Together these have a substantial influence on the ability to detect a fracture.

Resolution (spatial resolution) is the ability of an imaging system to depict two objects as separate as these get smaller and closer together, i.e. how small an object can be seen on a given modality [1]. Higher spatial resolution is the ability to see objects that are smaller and closer together. The historical method of measuring radiographic and CT spatial resolution was by using test phantoms that actually measured the ability to separate line pairs per millimetre. Many factors influence spatial resolution for all modalities. Most importantly, in digital imaging systems is the pixel size. The size of the pixel is determined by the number of pixels across the field of view. Thus, a larger field of view with the same pixel matrix will result in lower resolution. Objects smaller than the pixel size cannot be resolved as separate structures. Blurring in the image will also detract from spatial resolution, thus geometric magnification and motion (patient or imaging apparatus) should be avoided. In cross‐sectional imaging, in plane resolution is directly related to pixel size, but the z‐axis (slice thickness) determines the voxel size. In cross‐sectional modalities, the z‐axis is an important consideration in the identification of fractures. If a linear structure or plane such as a fracture is oblique to the acquisition plane, the margins of the line/plane will be blurred by a factor related to the slice thickness (z‐axis/voxel size) and the angle of obliquity through the image. Spatial resolution is particularly important in the diagnosis of incomplete or non‐displaced fractures. For many modalities, the disruption of mineral substance in these cases will be at the limits of spatial resolution.

Image noise is an important contributor to degradation of image quality or degradation of the utility of a given image. Noise caused by various systematic or random variables contributes extraneous optical density (echogenicity, signal intensity, etc.). Digital imaging systems (as compared to film screen systems) have systematic noise from the electronics and the structure of the detector. Anatomic structures that are not of interest to the viewer are also a form of noise, e.g. radiographs with bowel superimposed over the lumbar vertebral bodies. Quantum noise is important in digital diagnostic imaging. In most instances, images made with X‐rays and gamma rays use the lowest number of rays (quanta) possible to obtain a diagnostic image. When the dose is limited, the ratio of useful to non‐useful information (noise) shifts in favour of the latter. Thus, increasing the dose of radiation can shift the ratio towards useful information. In MRI, this is achieved by recording intensity multiple times (number of excitations).

Bit depth determines the number of possible shades of grey that can be applied to the imaging systems output. Most medical imaging devices range from 10 to 14 bit depth thus having the capability of recording 1024, 4096, or 16 384 shades of grey. This is beyond the limits of most digital displays and human resolution. The conversion of the image from a 10 bit depth image (1024) shades of grey to something more useable occurs by the application of a lookup table that determines the displayed greyscale values relative to the recorded greyscale value.

Contrast‐to‐noise ratio (CNR) and signal‐to‐noise‐ratio (SNR) are computations that describe the relationship between the important image quality parameters and the noise of the image. CNR computes the difference in signal between the object and background, divided by the background noise. SNR computes the integrated signal of the object (pixel signal minus noise), on a per pixel basis, independent of size and homogeneity, divided by the background noise. The SNR is a useful metric that is closely related to lesion conspicuity or the observer's ability to detect the lesion. Even without a numerical (computed) value for SNR, it becomes a visual cue to experienced observers whereby image degradation due to a low SNR is readily evident (Figure 5.1).

Accuracy and precision are necessary for reliable interpretation of images. These, in turn, depend not only on image quality but also the ability and experience of the observer to identify true positives and true negatives.