| Περίληψη: | The number of patients who have undergone some kind of internal fixation or joint replacement is increasing thanks to the development of technology and orthopaedics. All these patients carry metal implants. Magnetic resonance imaging has an advantage over other imaging methods, due to its superior soft tissue contrast and to its sensitivity in detecting the inflammation which is present at infections and malignancies. However metal implants usually deteriorate the image quality and as a result affect the accuracy of the diagnostic procedure. This is the case when the region of interest is in the proximal vicinity of the implant, or the implant is large enough. A number of MRI sequences have been proposed in order to overcome the artifact that comes from metal implants, more formally known as susceptibility artifact. However the most effective of them, are not widely available. The need for optimization of MR imaging at the presence of metal implants presupposes the development of methods capable of quantifying the artifact under various imaging sequences and conditions.
Most artifact quantification techniques proposed until now, are usually based on the visual observation (experienced radiologists) or at image segmentation methods. These segmentation methods, segment the image based on arbitrary selected gray values (thresholds). A more objective and precise quantification method relies on the subtraction of images of a zero artifact replica (test object) from those of the real metal implant. The copy is constructed from material with similar values of magnetic susceptibility with its environment (usually water). The images deriving from the copy if we take in consideration the noise differences, have no susceptibility artifact. In this method artifact is quantified as energy differences between the two images [Kolind S et al, 2004]. Since the acquisition conditions are identical except the presence of susceptibility artifact in the image depicting the real metal object, the energy difference is used to quantify the artifact. While the method quantifies the artifact, giving precise values, it does not inform us for its position in space
At this thesis we proposed a new, to our knowledge, method of artifact quantification. It is based in the physical cause of the artifact, which are the gradients of the magnetic field, which derive from the presence of the metal implant. The gradients of the magnetic field create corresponding gradients at the gray scale values of the image. These gradients may be detected if we apply suitable filter which detects the amplitude of the gradient. In this way we detect both regions with signal void (low signal intensity) and signal pill ups (high signal intensity). That means that we do not have to apply two different operators to segment two regions of the artifact with so different signal intensity values. Then the image is thresholded using a fully automated algorithm, proposed by [Li & Lee 1993]. This algorithm is available in image analysis environment ImageJ.
At the first part of this thesis there are presented the basic principles of nuclear magnetic imaging image formation. The interaction of the most common materials with the magnetic field is also presented. All these are considered necessary to explain the generation of magnetic susceptibility artifact at the image acquired. The theory beyond the magnetic susceptibility artifact generation is then explained in detail.
At the experimental part of this thesis, the proposed algorithm is applied to the imaging of two implants (made of titanium and antimagnetic stainless steel) at the sequences which are most commonly used to musculoskeletal MRI. The proposed algorithm is compared with a variation of the method of the image energy differences proposed by [Kolind Sh, 2004]. This method quantifies the artifact as energy difference of image of the real implant from the image of a replica with zero susceptibility artifact (reference image). In the present thesis the image of lower susceptibility artifact (obtained at higher bandwidth) is considered as reference image. In our case it is assumed that the energy difference among different bandwidth acquisitions is negligible in relation to the susceptibility artifact amplitude. This assumption allows as to use instead of energy differences, the differences in the gray scale values of the image instead. Statistical analysis showed moderate to strong positive correlation between the two methods. Possible reasons of not obtaining strong correlation at all measurements is due to the regions of the image that the proposed algorithm quantifies. By segmenting regions of high gradient, we focus mainly at regions where there is high variation at the gray scale values. However, in many cases nearly homogeneous regions of an image, with little or no alteration in gray scale values, may also be considered as artifact. These areas are not segmented as artifact when the proposed algorithm is applied. More over the assumption of considering negligible the noise contribution between the different acquisitions may be an oversimplification.
Nevertheless, the proposed algorithm, is an objective repeatable and observer independent method. Moreover it is capable of determining the boundaries of the artifact in image space. It is not intended to be used as a method of absolute quantification of the susceptibility artifact. It should be used as means of comparison of acquisitions concerning the same sequence. Its combination with an additional algorithmic step, such as one which detects image features may result in a powerful tool of image artifact quantification. This more sophisticated version of this proposed algorithm should be adequate enough to quantify the artifact not only at phantom models but even at the everyday clinical practice.
|
|---|
