By generating a synthetic CT image (sCT) directly from the MRI, the electron density can be calculated, and the CT examination can be excluded from the patient flow minimizing the risk of uncertainties in the registration. Basing the radiation treatment process solely on MR images is called MRI-only and is beneficial for the patient as it can provide more accurate radiation treatment than the standardised treatment with fewer CT examinations and possibly a more cost-effective radiation treatment process.  The conventional coils that are normally used in MRI for dose planning purpose cannot be placed directly on the patient as the outer body contour then can be deformed by these relatively heavy coils. The coils are therefore placed on a special holder which creates distance between the coil and the patient, this degrades the signal to noise ratio (SNR). The department for radiation treatment at Linköping University Hospital has access to a newly developed coil with so-called Air Technology. This type of coil is significantly lighter than the conventional ones and the idea is that this coil can be placed directly on the patient without causing deformation.  The aim of this project is to develop a software tool to validate an MRI-only workflow and to investigate if the radiation dose calculation based on sCT data differs from calculations based on CT data. Furthermore, to examine if the AIR coil has any effect on the body contour and the calculated dose.  For the evaluation of the AIR coil three similarity comparison methods were used, Hausdorff distance, Dice similarity coefficient (DSC) and Surface DSC. The result for the Hausdorff distance showed that eight out of eleven comparisons were within 4 mm difference, this corresponded good with Surface DSC where eight out of eleven had a result over 99% at a 3 mm tolerance. DSC measures gave above 98.5% for nine out of eleven of the comparisons.  The investigation on whether the radiation dose calculation differed was done using the dose- volume histogram statistics in Eclipse. A method calculating the gamma index was implemented in MICE. The results showed that nine out of ten gamma indexes had deviations that were within the same range. An explanation for why the results of one patient were not within the same range as the others could not be found and needs further investigations.

No description available.

La tomodensitométrie à faisceau conique (CBCT) est répandue à travers les centres de cancérologie pour le positionnement du patient avant chaque traitement de radiothérapie. Ces images pourraient potentiellement être utilisées pour des tâches plus complexes. En radio-oncologie, les techniques de planification adaptative sont en développement et nécessitent l'utilisation des images de tomodensitométrie synthétiques (sCT). Alors, plusieurs groupes de recherche ont proposé différentes techniques pour générer des images sCT à partir des données CBCT. L'objectif principale de ce projet est d'évaluer une nouvelle méthode d'apprentissage profond pour générer des images sCT de pelvis à partir des images CBCT. Onze patients ont été rétrospectivement étudiés. Chaque patient a été imagé en séquence par deux techniques d'imagerie volumétrique dans la même position, soit un scan au CBCT et un autre au CT sur rails (CTr). Afin de pouvoir utiliser les images synthétiques dans un contexte clinique, la qualité de l'image et l'impact dosimétrique entre les sCT et les CTr doivent être évalués. Les nombres CT des images sCT sont comparés à ceux des images CTr. Finalement, pour évaluer l'impact dosimétrique, les plans de traitement optimaux sont recalculés sur les images sCT et CTr. Les différences de dose sont évaluées à l'aide d'une analyse gamma et des histogrammes dose-volume. L'évaluation quantitative montre qu'il y a des différences statistiquement significatives dans les os et les cavités d'air. Tandis que, les différences des tissus adipeux et mous ne sont pas statistiquement significatives. Les doses estimées dans les organes à risque et les PTV à partir des données des sCT sont surestimées comparativement à celles calculées à partir des données des CTr. Cependant, les erreurs de doses sont inférieures à 2% pour la plupart des cas étudiés. Ces erreurs de doses sont probablement causées par le manque de tissus dans la périphérie du patient et les erreurs des nombres CT. Bien que les différences de doses soient cliniquement acceptable, la méthode proposée devrait temporairement être limitée aux validations quotidiennes de plans de traitement pour des cas pelviens. / The cone-beam computed tomography (CBCT) is widely spread in cancer centers for positioning the patient before their radiotherapy treatment. These images could potentially be used for more complex tasks. In radio-oncology, adaptive planning technics are in development and require the use of synthetic CT (sCT) images. So, multiple research groups proposed different methods to generate sCT images from CBCT data. The main purpose of this project is to assess a new deep-learning method to generate sCT images from CBCT images. Eleven patients were retrospectively studied. Each patient was subsequently imaged by two volumetric imaging methods in the same position, one CBCT scan and the other from CT on rails (CTr). In order to clinically use the synthetic images, image quality and dosimetric impact between sCT and CTr must be evaluated. sCT images' CT values are compared to those in CTr images. Finally, to evaluate the dosimetric impact, optimal treatment plans are recalculated with sCT and CTr images. Dose differences are assessed by gamma analysis and dose-volume histograms. The quantitative evaluation shows that differences are statistically different in bones and air cavities. As for adipose and soft tissues, differences were not statistically different. The estimated doses in organs-at-risk and PTVs from sCT data are overestimated compared to those from CTr data. However, dose errors are inferior to 2% in the majority of studied cases. These dose errors are most likely due to missing tissues on the outskirt of the patient and the errors of CT numbers. Although dose differences are clinically acceptable, the proposed method should temporarily be limited to daily validations of pelvic treatment plans.

Magnetic resonance (MR) images together with computed tomography (CT) images are used in many medical practices, such as radiation therapy. To capture those images, patients have to undergo two separate scans: one for the MR image, which involves using strong magnetic fields, and one for the CT image which involves using radiation (x-rays). Another approach is to generate synthetic CT (sCT) images from MR images, thus the patients only have to take one image (the MR image), making the whole process easier and more effcient. One way of generating sCT images is by using generative diffusion models which are a relatively new class in generative models. To this end, this project aims to enquire whether generative diffusion models are capable of generating viable and realistic sCT images from MR images. Firstly, a denoising diffusion probabilistic model (DDPM) with a U-Net backbone neural network is implemented and tested on the MNIST dataset, then it is implemented on a pelvis dataset consisting of 41600 pairs of images, where each pair is made up of an MR image with its respective CT image. The MR images were added at each sampling step in order to condition the sampled sCT images on the MR images. After successful implementation and training, the developed diffusion model got a Fréchet inception distance (FID) score of 14.45, and performed as good as the current state-of-the-art model without any major optimizations to the hyperparameters or to the model itself. The results are very promising and demonstrate the capabilities of this new generative modelling framework.

Both computed tomography (CT) and magnetic resonance imaging (MRI) have a purpose for radiotherapy. But having two imaging sessions brings uncertainty which makes it beneficial to create synthetic CT (sCT) images from MR images. In this work a Generative Adversarial Network (GAN) was designed and implemented for sCT generation. The purpose of the work was to broaden the understanding of how variation in training data affects a model’s performance on generating sCTs. This was done by increasing the training sets with patients with artifacts, female patients and synthetic MR contrasts. Eight different machine learning models with varying training data were trained and evaluated. Four models were trained using T2-weighted data only while the other four used both real T2-weighted images and synthetic T1 (sT1) images in their training sets. The models were evaluated on the pixel value difference between the CT and the resulting sCTs using a mean absolute error (MAE) evaluation. Afterwards, dose calculations were made with patients’ treatment plans on both their CTs and their corresponding sCTs and compared doses to some of the structures. Finally the models were compared based on their performance on synthetic MR contrasts. This means I used a contrast transfer model to change the contrast from a T1-weighted image to a synthetic T2 or from a T2-weighted image to a synthetic T1 and then generated sCT images from the synthetic contrasts. These experiments showed that when I increased the models’ training sets with relevant patients MAE decreased between the CT and a generated sCT. Importantly, this was also true for our models trained on sT1s when evaluating on T1 weighted images. Increasing the size of the datasets also increased the performance in a treatment planning purpose and it also decreased the difference when evaluating the models on original MR images and synthetic MR images. In conclusion an improved performance was shown for models evaluated on images with artifacts, female patients and other MR contrast when including more images from those image types in the training dataset.