Magnetic resonance imaging in radiotherapy treatment planning

Abstract

From its inception in the early 1970's up to the present, magnetic resonance imaging (MRI) has evolved into a sophisticated technique, which has aroused considerable interest in var- ious subelds of medicine including radiotherapy. MRI is capable of imaging in any plane and does not use ionizing radiation by virtue of which MRI lends itself admirably to the purpose of prolonged time course studies. MRI is capable of excellent spatial resolution and it presents information over large areas of the body. The MRI signal depends on multiple parameters resulting in excellent contrast resolution of the soft tissues. Within the realm of radiotherapy, MRI oers prospects with regards to identication of tissues and tissue ab- normalities (tumour, oedema, necrosis, brosis, cysts), determination of tumour extent in relation to surrounding tissues and organs and assessment of response to treatment. The goal of radiotherapy is to administer a high dose to the tumour while sparing healthy sur- rounding tissues as much as possible. Since tight margins around the target are applied, accurate information on tumour extent is of great value in radiotherapy treatment planning (RTP). Also information on motion of the tumour and surrounding organs, e.g. caused by respiration, is of importance for dening the margins and can be acquired by fast MR imaging techniques. However, the introduction of MRI into RTP is seriously hampered by geometry and intensity distortions which are known to be present in MRI. These distortions are caused by non-idealities of the equipment (non-uniformity of the static magnetic eld and non-linearities of the gradient magnetic elds of the MRI scanner) and by magnetic eld perturbations induced by the object to be imaged, in this case the patient. These magnetic eld inhomogeneities and gradient eld non-linearities lead to image distortions, the severity of which depends on the type of pulse sequence and its parameters. The aim of this thesis is to investigate the capabilities of MRI in radiotherapy treatment planning and to explore the MR image distortions and how distortions can be reduced or, if necessary, corrected in order to integrate MRI into RTP in a reliable manner. Image distortions and the ecacy of correction methods were evaluated in phantom, volunteer, and patient studies. Furthermore, the potential of MRI in organ motion studies was investigated and it was explored whether the image artifacts induced by I-125 seeds could be used to evaluate permanent prostate implants in brachytherapy. Chapter 1 introduces the potentials and problems with regard to the use of MRI for ra- diotherapy treatment planning and gives a review of the literature concerning MR image distortions, integration of MRI into RTP, MRI organ motion studies for RTP, and MRI- based brachytherapy evaluation. Chapter 2 brie y reviews the basic principles of nuclear magnetic resonance (NMR), spatial encoding in MRI, and the sources of geometry and intensity distortions in MRI, viz. machine- related magnetic eld inhomogeneity and gradient non-linearity and patient-related magnetic eld inhomogeneity due to chemical shift and susceptibility. Chapter 3 describes the measurement, analysis, and correction of machine dependent geo- metric distortions in MRI with special attention for phantom design and eld error stability in time and for dierent pulse sequence parameters. Inhomogeneity of the static eld and non-linearity of the gradients was established by phantom experiments. A grid phantom of equally spaced tubes appeared to be very suitable for this purpose. Interchanging the directions of the read-out and the phase-encoding gradients enabled decomposition of the image distortions into contributions from static eld inhomogeneity and the non-linearity of the three gradients. A 3D map of static eld inhomogeneity and non-linearity of the gra- 112?dients was thus obtained from sagittal, coronal, and transversal multiple slice images with for each acquisition the phantom positioned such that the tubes were perpendicular to the image plane. Time series of measurements on the Gyroscan S15 showed eld error stability within the experimental errors of 1 ppm for static eld inhomogeneity and 1 mm for the gradient elds. Measurements on the Gyroscan ACS-NT, equipped with active shielding technology, showed eld error stability under dierent imaging conditions. These observa- tions imply that the measured error maps can be used for correction of patient images which may have been acquired with a pulse sequence that is not necessarily identical to the pulse sequence applied for phantom imaging. The correction procedure reduced distortions up to 13 mm within a volume of interest (VOI) with dimensions 336 x 336 x 210 mm 3 to smaller than 2 mm. In a study on the Leksell frame, distortions up to 6.4 mm were reduced to smaller than 1.5 mm. We conclude that MR image corrections are necessary in applications which require mm accuracy and that correction methods, based on 3D maps of static eld inhomogeneity and gradient non-linearity, are feasible in clinical practice. Chapter 4 describes the analysis of the patient related magnetic eld perturbations and resulting image distortions in case of MRI of the head and the pelvic region. The magnetic eld was calculated by numerically solving the Maxwell equations for a magnetostatic eld. The magnetic eld around the head resembled a dipole eld in the midsagittal plane of calculation with minimal eld perturbation on the diagonals. The magnetic eld in the localization frame depended strongly on the orientation of the Perspex plates with respect to the applied magnetic eld. The maximum spatial distortion of external landmarks in the localization frame amounted 12.8 mm in a 1.5 T MR head image acquired with a relatively weak read-out gradient of 0.68 mT/m. Susceptibility-induced distortions in the pelvic region were smaller than 3 mm in 1.5 T images acquired with a read-out gradient strength of 0.54 mT/m. Since susceptibility induced distortions are proportional to the static magnetic eld strength and inversely proportional to the gradient strength, we may conclude that object-related distortions can be reduced to the order of the pixel size by imaging at 0.5 T and using gradient strengths on the order of 3 mT/m. Chapter 5 describes qualitatively the in uence of the type of pulse sequence and its param- eters on geometry and intensity distortions. The strength of the read-out gradient, which is controlled by the water fat shift parameter, was the main factor aecting the severity of the susceptibility artifact. The range of water fat shifts, that could be selected, was in uenced by the eld of view and the acquisition matrix. In echo planar imaging (EPI), the EPI factor (number of proles acquired after a single excitation) strongly in uenced the water fat shift (in pixels) in the phase encoding direction. In applications of MRI, which require geometric accuracy, we should be aware that these parameters indirectly aect the severity of image distortions. Spatial distortions occurred in the direction of the read-out gradient in spin echo (SE) and fast eld echo (FFE) imaging, but also and more strongly in the direction of the phase-encoding gradient in echo planar imaging. The direction of the read-out gradient was controlled by the fold-over direction parameter which is the direction of the phase-encoding gradient. Signal loss was most severe in FFE images acquired with relatively long echo times and/or in case of partial k -space sampling (reduced scan matrix, half matrix or partial echo). Generally, highest accuracy and least signal loss is achieved in spin echo imaging with minimal water fat shift and in fast eld echo imaging with minimal water fat shift and short echo time. 113?Chapter 6 describes the investigation of geometric distortions in 1.5 T MR images for use in radiotherapy treatment planning of patients with brain tumours. Patients underwent magnetic resonance imaging in the radiotherapy position with the head xed by a plastic cast in a Perspex localization frame. For purposes of accuracy assessment, external and internal landmarks were indicated. Tubes attached to the cast and in the localization frame served as external landmarks. In the mid-sagittal plane the brain-sinus sphenoidalis interface, the pituitary gland-sinus sphenoidalis interface, the sphenoid bone and the corpora of the cervical vertebra served as internal landmarks. Landmark displacements as observed in the reversed read-out gradient experiments were analyzed with respect to the contributions of machine related static magnetic eld inhomogeneity and susceptibility and chemical shift artifacts. In this study at 1.5 T with read-out gradient strength of 3 mT/m, machine related and susceptibility induced static magnetic eld inhomogeneity were on the same order, resulting in spatial distortions between