Imaging and radiotherapy physics topics for project and master thesis

Transcription

1 Imaging and radiotherapy physics topics for project and master thesis Supervisors: Assoc. Professor Kathrine Røe Redalen and PhD candidates Franziska Knuth and Kajsa Fridström. Contact: office D4-164 There has been a huge expansion in the use of medical imaging in cancer diagnostics, and treatment evaluation, where particularly magnetic resonance imaging (MRI) is being increasingly used due to the method s ability to provide high-resolution images without the use of ionizing radiation. Anatomical MRI is important in the diagnostic work-up in many cancer types, where it allows visualisation of zonal anatomy and detailed evaluation of local disease extent, regional metastasis and general anatomy. Further, functional MRI comprises sequences reflecting functional / biological tissue properties, such as tissue structure and cell density (diffusion-weighted MRI), metabolism (MR spectroscopy) and various features of tumour vascularity (dynamic contrast-enhanced (DCE) MRI and dynamic susceptibility contrast (DSC) MRI). These techniques have shown potential to also capture important features of the tumour microenvironment relevant for radiotherapy efficacy. We are investigating how these techniques may be optimally utilised in order to provide pre-treatment measures of whole-tumour radiosensitivity that may be aid planning of individualised radiotherapy strategies. In addition, we are investigation how these techniques may be used both for early monitoring and end-oftherapy evaluation of radiotherapy outcome. Figure 1: Functional MR images of rectal cancer acquired on a 1.5 T MR scanner. (A) T2-weighted morphologic image, (B) diffusion-weighted (DW) image with b-value of 500 s/mm2, (C) Parametric bmax map from DW image overlaid the T2-weighted image, (D) multi-echo T2*-weighted image, (E) T1-weighted image before injection of contrast agent, (F) T1-weighted image after injection of contrast agent, (G) Parametric blood perfusion map from dynamic contrast-enhanced image overlaid the T2-weighted image, (H) Spectra from in vivo MR spectroscopy.

2 In a study entitled OxyTarget Functional MRI of hypoxia-mediated rectal cancer aggressiveness functional MR data from 180 patients are collected to investigate whether functional MR images (or a combination of several images multiparametric MRI) can identify the radiosensitivity of tumours. The study also includes collection of histology/immunohistochemistry of surgical specimens, clinicopathologic outcome parameters, radiotherapy plans and patient survival which allows broad crossinvestigations as well as interpretation of the biology behind the MR images. Example functional MR images from a patient in this study are shown in Figure 1. Below are presented some proposals for project- and master theses. Candidates may also take contact for discussion of alternative proposals. Proposal 1: DW-MRI and DCE MRI for tumour perfusion imaging Estimates on tumour blood perfusion is commonly acquired by voxel-wise kinetic modelling of contrast-enhancement curves from T1-weighted DCE MRI, which requires injection of a contrast agent. We have in a small pilot study recently shown that alternative estimates on tumour blood perfusion may be acquired through so-called intravoxel incoherent modelling (IVIM) of diffusion-weighted (DW) MRI data (Bakke et al, 2017), which does not require injection of a contrast agent, and which were shown to both predict radiotherapy response and patient survival. Figure 2: Map of kep, a perfusion marker estimated from DCE MRI. Made by project student in 2017, Tina S. Evensen. The image shows that perfusion is heterogeneous in the tumour. In this proposal, the candidate will use Matlab (or a similar tool) to estimate blood perfusion parameters from both DW MRI and DCE MRI and thereafter perform immunohistochemical analysis of tumour perfusion in tissue (the surgical specimen) in order to explain the MRI findings. We have surgical specimens stained with the endothelial marker CD34 available for this project. Proposal 2: T2*-weighted MR imaging for hypoxia detection T2* relaxation refers to decay of transverse magnetisation caused by spin-spin relaxation or magnetic field inhomogeneity. T2*-weighted MR images may be acquired by gradient echo MRI with multiple echo-times (TE) or by DSC MRI. The inverse T2* is R2* and is in

3 theory a measure of the oxygen saturation in the tissue. Oxygen (or oxygen deficiency; hypoxia) is an important determinant for radiotherapy response and outcome in patients with solid cancer. Figure 3: An R2* peak tumour map from one patient with rectal cancer. This figure was part of our publication in Journal of Magnetic Resonance Imaging (by E. Grøvik et al, 2017;46(1): ). In this study, we found that the R2* peak could predict whether the patient had benign or malignant lymph nodes, which is an important prognostic factor for patient outcome. In this proposal, the candidate is expected to use Matlab (or a similar tool) to estimate R2* tumour maps from both multi-te MRI and DSC MRI data acquired both before and after chemoradiotherapy. Thereafter, the candidate will evaluate the potential of these measures to assess tumour radiosensitivity and their correlation to tumour hypoxia by comparison with immunohistochemical analysis of tumour hypoxia in tissue. We specimens stained with the hypoxia-inducible factor 1 alpha (HIF-1 alpha) and carbonic anhydrase (CAIX). Depending on the candidate s competences and interests the generated MR data may also be further used to simulate various approaches of delivering radiotherapy based on the R2* tumour maps, for instance through dosepainting or radiation boosting techniques. Proposal 3: MRI-based texture analysis It is acknowledged that solid tumours are heterogeneous and that mean tumour values of estimated MR parameters do not reflect the heterogeneous information contained in the tumours. One way to capture tumour heterogeneity is through texture analysis. The term texture refers to specific properties of the internal structure of image regions, for example rough versus smooth or oriented versus randomly dispersed (among others). In medical image analysis, texture-based methods are very useful to classify tissue types, typically by first extracting texture features and then feeding these features into a classifier. However, investigations on the potential of texture-based methods of imaging data in the context of radiotherapy is poorly investigated, and particularly the potential value of texture-based analysis of functional imaging data in the context of radiotherapy. In this proposal, the candidate will use various functional MR images from the OxyTarget data set and explore how texture analysis may provide added value in detection of tumour radiosensitivity and radiotherapy outcome compared to mean parameter values. The candidate will start with lower-order histogram features (such as kurtosis, entropy and skewness) and may proceed to more advanced texture analysis

4 based on for instance local binary patterns (BNP), gray-level co-occurrence matrix (GLCM), filtering approaches (e.g. Gabor filters) and various classifiers and multi-channel visualisation methods. Proposal 4: Automated image segmentation for radiotherapy Due to rapid advances in radiation therapy, especially image-guidance and treatment adaptation, a fast and accurate segmentation of medical images is a very important part of the treatment. Manual delineation of target volumes and organs at risk is still the standard routine, even though it is time-consuming and prone to intra- and interobserver variations. Automated segmentation methods seek to exploit machine-learning algorithms to reduce delineation workload and unify the organ boundary definition. In radiotherapy, CT is currently the most common imaging modality, however, there is a rapid shift to more frequent use of MRI, which provides superior soft tissue contrast. This is particularly relevant also due to the introduction of the combined MRI-linear accelerator (MRI-Linac). The improved soft tissue contrast provided by MRI enables better identification of various tissue structures, which may enable improved performance of segmentation algorithms. Figure 4: Rectal tumour delineated on anatomical T2-weighted MR image. The illustration was made by project student in 2017, Tina S. Evensen. In this project, we aim to explore different MRI-based segmentation methods based on machine-learning and by combining anatomical with functional MR images to see what approaches performs better. We will compare the results to tumor delineations already made by two independent radiologists. Proposal 5: MR-based radiotherapy dose-painting: modelling studies Radiotherapy dose-painting strategies relies on the use of medical images that reflects radiobiological tumour properties. These images are used to prescribe an inhomogeneous tumour dose in order to increase the local control compared to conventional radiotherapy with a homogenous tumour dose. Two dose-painting approaches exist: dose-painting by contours (DPBC) and dose-painting by numbers (DPBN). In DPBC the most aggressive sub-volume within the tumour is delineated and subsequently dose-escalated to a homogenous dose, while in DPBN each tumour voxel is given a dose according to the voxel value in the corresponding biological image.

5 Figure 5: Left: example standard radiotherapy plan, right; illustration of the concept of dosepainting by increasing the dose > 2 Gy in some voxels and decreasing the dose > 2 Gy in other voxels. This strategy relies on functional imaging of radiobiological properties. In our projects we investigate whether functional MRI is useful for this purpose. In rectal cancer, DPBC may be an alternative in two different scenarios: 1) to increase the probability of obtaining a better tumour response to preoperative chemoradiotherapy in aggressive, poor-responding tumours in order to enable subsequent surgery, and 2) to increase the probability of obtaining a complete response to chemoradiotherapy and thereby omit the surgery to achieve an improved quality of life with sparing of normal tissue structures (no colostomy). This project includes modelling studies where you will estimate the tumour control probability (TCP) based on utilising MR-data to plan the DPBC for one or both of the situations 1 and 2 above. Standard TCP models assume uniform tumour cell density across the tumour. Our aim is to develop an individualised TCP model by including functional MR information. MR-data that can be used is apparent diffusion coefficient (ADC) maps from DWI MRI to extract tumour cell density distributions and R2* maps from T2*-weighted MRI to extract hypoxia distributions across the tumour. The number of cells / initial amount of hypoxia within imaging voxel can be explored by different approaches by using the ADX maps; linear, binary and sigmoid relations. The TCP models will be based on linear-quadratic (LQ) cell survival curves with appropriate alpha/beta values. The aim is to develop a method that can be evaluated prospectively in clinical studies. Proposal 6: Quantitative tumour histology analysis The purpose of this project is to describe the intratumoural network through image analysis of histological sections. This information is useful both to describe the aggressiveness of a tumour and to understand how therapies have or will work. The tumour vasculature is characterized by complexity, irregularities and poorly regulated growth. Fractal analysis has been used to establish that tumour vasculature has a different network architecture from that of the normal arteriovenous system or the capillary network. The vasculature is responsible for the transportation of oxygen to tumour cells. Hypoxia (oxygen deficiency) is a challenge in treatment of cancer, both through its indirect biological effects, such as on cell cycle progression, but also through direct chemical effects. In particular, the oxygen effect reduces the effectiveness of radiation

6 therapy. Furthermore, the network morphology relates to many other parameters as well, such as the angiogenic and the metastatic capability of the cancer. This raises the possibility of using image analysis, and fractal analysis in particular, to quantify different aspects of the network morphology. In this project, tissue sections stained with the endothelial marker CD34 as well as with the hypoxia markers HIF-1 alpha and CAIX will be available. The impact of the immune system and its response may also be included through stains with the markers CD3 and CD8. You will quantitatively assess parameters mainly through fractal analysis and syntactic structure analysis. Some parameters, such as the number of vessels, the size of the vessels, the total vascular area, and cumulative histograms of distances to the nearest vessel, can be obtained directly from the images. The results will be compared with the patients clinical outcome data in order to also identify whether some of the parameters are associated to diagnosis, treatment outcome or survival. Figure 6: CD34 images depicting the endothelium (blood vessels) of two different tumours. We can see different vessel size, area coverages and micro vessel densities (MVD). Figure 7: Illustration of a potential methodology to quantify vascular features. A) Individual vessel characterisation: endothelial area (red), luminal area (blue), vessel/vascular area (red + blue), vessel perimeter (stapled green), width (diameter of imposed yellow circle), skeleton (black line), branching points (yellow markers), convex hull (interior of stapled black line, i.e. vessel + gray). B) Contextual vasculature characterisation: centre of vessel (yellow markers), neighbouring vessels according to the Gabriel s Graph criterion (blue lines), i.e. all connections between vessels such that no other vessel centres are found within the circle spanned by the connecting line (see blue circle). Both figures are from Mikalsen et al., 2013.