Medical image analysis is a ubiquitous and essential part of modem health care. A crucial first step to this is segmentation, which is often complicated by many factors including subject diversity, pathology, noise corruption, and poor image resolution. Traditionally, manual tracing by experts was done. While considered accurate, this process is time consuming and tedious, especially when performed slice-by-slice on three-dimensional (3D) images over large datasets or on two-dimensional (2D) but topologically complicated images such as a retinography. On the other hand, fully-automated methods are typically faster, but work best with data-dependent, carefully tuned parameters and still require user validation and refinement. This thesis contributes to the field of medical image segmentation by proposing a highly-automated, interactive approach that effectively merges user knowledge and efficient computing. To this end, our work focuses on graph-based methods and offer globally optimal solutions. First, we present a novel method for 3D segmentation based on a 3D Livewire approach. This approach is an extension of the 2D Livewire framework, and this method is capable of handling objects with large protrusions, concavities, branching, and complex arbitrary topologies. Second, we propose a method for efficiently segmenting 2D vascular networks, called ‘Live-Vessel’. Live-Vessel simultaneously extracts vessel centrelines and boundary points, and globally optimizes over both spatial variables and vessel radius. Both of our proposed methods are validated on synthetic data, real medical data, and are shown to be highly reproducible, accurate, and efficient. Also, they were shown to be resilient to high amounts of noise and insensitive to internal parameterization. / Applied Science, Faculty of / Electrical and Computer Engineering, Department of / Graduate

No description available.

No description available.

Background To combat one of the leading causes of death worldwide, lung cancer treatment techniques and modalities are advancing, and the treatment options are becoming increasingly individualized. Modern cancer treatment includes the option for the patient to be treated with proton therapy, which can in some cases spare healthy tissue from excessive dose better than conventional photon radiotherapy. However, to assess the benefit of proton therapy compared to photon therapy, it is necessary to make both treatment plans to get information about the Tumour Control Probability (TCP) and the Normal Tissue Complication Probability (NTCP). This requires excessive treatment planning time and increases the workload for planners.  Aim This project aims to investigate the possibility for automated prediction of the treatment dose distribution using a deep learning network for lung cancer patients treated with photon radiotherapy. This is an initial step towards decreasing the overall planning time and would allow for efficient estimation of the NTCP for each treatment plan and lower the workload of treatment planning technicians. The purpose of the current work was also to understand which features of the input data and training specifics were essential for producing accurate predictions.  Methods Three different deep learning networks were developed to assess the difference in performance based on the complexity of the input for the network. The deep learning models were applied for predictions of the dose distribution of lung cancer treatment and used data from 95 patient treatments. The networks were trained with a U-net architecture using input data from the planning Computed Tomography (CT) and volume contours to produce an output of the dose distribution of the same image size. The network performance was evaluated based on the error of the predicted mean dose to Organs At Risk (OAR) as well as the shape of the predicted Dose-Volume Histogram (DVH) and individual dose distributions.  Results  The optimal input combination was the CT scan and lung, mediastinum envelope and Planning Target Volume (PTV) contours. The model predictions showed a homogenous dose distribution over the PTV with a steep fall-off seen in the DVH. However, the dose distributions had a blurred appearance and the predictions of the doses to the OARs were therefore not as accurate as of the doses to the PTV compared to the manual treatment plans. The performance of the network trained with the Houndsfield Unit input of the CT scan had similar performance as the network trained without it.  Conclusions As one of the novel attempts to assess the potential for a deep learning-based prediction model for the dose distribution based on minimal input, this study shows promising results. To develop this kind of model further a larger data set would be needed and the training method could be expanded as a generative adversarial network or as a more developed U-net network.

Background. Classification based PET/MRI attenuation correction methods utilize a PET/CT derived method of identifying tissues and assigning them attenuation values. Phantoms for PET/MRI generally use theoretical attenuation maps.Aim. This paper investigates the creation of fluids that can be used in phantoms for PET, CT, and MRI. The focus lies on creating properties that result in the attenuation to be correctly estimated with CT and MRI systems. This is a step toward the development of hybrid phantoms with tissue equivalency in all modalities.Method. Attenuation and fat fraction is altered in water to produce a fluid that generates correct attenuation values when measured with both CT and MRI. The fat fraction influences what type of tissue the fluid is classified as in the PET/MRI system used. In turn, the tissue type has an attenuation value associated with it. An emulsion of vegetable oil and water is investigated for the manipulation of fat fractions. The CT attenuation measurement in Hounsfield is modified by altering the density of the fluid. The concentration of sodium chloride is used to modify the Hounsfield value. In the end, the phantom of oil, water, salt solution, and emulsion is validated by adding PET-isotopes. The measured attenuation-corrected radioactivity concentration in PET/CT and PET/MRI are compared.Results. The results show that modification of salt concentration is an effective and easily applicable method of altering the Hounsfield attenuation value of a liquid. The use of emulsions results in expected classification based on fat fraction. Long- and short-term stability of the emulsions may be improved. Analysis of radioactivity concentration shows comparable values between PET/CT and PET/MRI proving that the concept works. / Bakgrund. Klassificeringsbaserad attenueringskorrektion i PET/MR utnyttjar en metod för PET/CT där vävnad identifieras och tilldelas ett attenueringsvärde. Fantomer för PET/MR använder vanligtvis teoretiska attenueringskartor.Mål. Den här rapporten undersöker skapandet av vätskor som kan användas i fantomer för PET, CT och MR. Fokuset ligger på att skapa egenskaper som ger samma korrekta resultat för attenuering i både CT- och MR-system. Detta är ett steg mot utvecklingen av hybridfantom med vävnadsekvivalens för alla modaliteter.Metod. Attenuering och fettfraktion ändras i vatten för att skapa en vätska som ger korrekta attenuringsvärden i både CT och MR. Fettfraktionen bestämmer vilken typ av vävnad som vätskan klassificeras som i det använda PET/MR-systemet. Vävnadsklassificeringen har ett associerat attenueringsvärde. En emulsion av vegetabilisk olja och vatten undersöks för manipulering av fettfraktioner. Attenueringsvärdet som mätt av en CT modifieras genom att ändra densiteten på vätskan. Koncentrationen av natriumklorid ändras för att påverka Hounsfield-attenueringsvärdet. Slutligen så undersöks fantomet med olja, vatten, saltlösning och en emulsion genom tillsatsen av PET-isotoper. Den attenueringskorrigerade radioaktiva koncentrationen som uppmättes i PET/CT och PET/MR jämförs.Resultat. Resultaten visar att modifikation av saltkoncentration är en effektiv och enkelt tillämpbar metod för att påverka Hounsfield-värdet på en vätska. Användningen av emulsioner leder till en förväntad klassificering baserad på fettfraktionen. Lång- och kortvarig stabilitet av emulsionerna kan förbättras. En analys av radioaktivitetskoncentrationen visar jämförbara värden mellan PET/CT och PET/MR vilket visar att konceptet fungerar.

Image quality has been found to be positively correlated with diagnosis accuracy. Radiologist aim for the highest quality image possible to determine the location of the suspected pathology. However, the most effective way of producing high quality images is to increase the radiation dosage to the patient. To avoid the many risks that come with radiation, patients want to keep dosage as low as possible. Diagnosing instruments are constantly being re-engineered and optimized to keep image quality high and radiation dosage low. If patients wish to avoid nuclear radiation exposure, alternative non-nuclear and low radiation modalities must be employed. The three most important metrics of image quality are spatial resolution, signal-to-noise (SNR) ratio and contrast-to-noise (CNR) ratio [1]. Radiologists and imaging technicians can do very little to improve the spatial resolution; and to improve the CNR a higher dosage is necessary to increase the value of every pixel. To increase radiation-SNR efficiency, the dosage can be reduced by 50% while only dropping the SNR by about 30% [2]. To simulate lower dosage, data is randomly taken out while the image is reconstructed until the acceptable SNR value is achieved. The broad applications can include reducing the signal-to-dosage ratio for any modality involving ionizing radiation and image reconstruction, reducing the risk for every imaged patient.

Non-invasive measurement of pH provides multiple potential benefits in oncology such as better identifying the type of drug that can be more effective in chemotherapy, potentially identifying tumors that are more likely to metastasize and also better assessing the treatment effects. Chemical Exchange Saturation Transfer (CEST) Magnetic Resonance Imaging (MRI) is a versatile non-invasive technique for molecular imaging. AcidoCEST MRI techniques have been developed over the recent years to perform tumor pH measurements by utilizing a contrast agent for which chemical exchange saturation transfer effects depend on the pH of the microenvironment. Quantitative description of CEST MRI signals are generally done via modeling Bloch-McConnell equations by incorporating pH as a parameter or by fitting Lorentzian line shapes to observed z-spectra and then computing a log ratio of the CEST effects from multiple labile protons of the same molecule (ratiometric method). Modeling using Bloch-McConnell equations is complicated and requires careful inclusion of many scan parameters to infer pH. The ratiometric method requires contrast agents that have multiple labile protons, thus making it unsuitable to use for molecules with a single labile proton. Furthermore, depending on the pH, sometimes it might not be possible to numerically compute the ratio due to the inability of detecting signal peaks for certain labile protons. Our aim here is to develop a machine learning based method that learns the CEST signal patterns from observed z-spectra on temperature and concentration-controlled contrast agent phantoms independent of the type of the contrast agent. Our results indicate that the machine learning method provides more general and accurate prediction of pH in comparison to the ratiometric method based on the phantom CEST dataset. Our method is more general in the sense that it does not require explicit modeling of signal peaks that are dependent on the type of contrast agent. We also describe a state of the art variational autoencoder based algorithm extending our machine learning method to measure tumor pH in vivo using AcidoCEST MRI on mouse tumor models.

No description available.

No description available.

The purpose of this experiment is to study as a model the gelation phase transition of agarose solutions with transverse relaxation (T2) quantitative magnetic resonance imaging (qMRI). The focus is on the reduction of T2 of agarose solution upon gelation. The sol-to-gel phase transition of agarose may provide a useful and controllable experimental model of tissue formation. Furthermore, it may provide the basis for exact mathematical models useful for understanding the much reduced transverse relaxation times (T2) observed in solid tissues relative to simple liquids. In this context, the purpose of this work was to monitor dynamically with T2 quantitative MRI the liquid-to-gel phase transition of pure agarose as a function of gel concentration. Samples of agarose at various concentrations were allowed to cool down while scanning dynamically with T2 qMRI, 32 x 10milliseconds (ms) echoes, CarrPurcell-Meiboom-Gill (CPMG), 3Tesla.T2 versus; (temperature).curves of each agarose solution show a distinct phase transition region characterized by a sharp T2 reduction. Four agarose solutions were sequentially prepared by dissolving agarose powder in distilled water at concentrations of 1%, 2%, 3%, and 4% by weight/volume. Immediately after preparation and boiling at 98°C, each liquid agarose solution was poured into a plastic container and scanned dynamically at 3.0T as it cooled down with a whole body MRI scanner (Achieva, Philips Medical Systems, Cleveland, OH). A single axial slice multi spin echo CPMG pulse sequence with the following parameters was used: 32 echoes, 10ms echo spacing, 1.5s repetition time (TR), 160 x 160 matrix size, and 2 SENSE factor. The time per dynamic scan was 1minute. The DICOM images were further processed with an adaptive T2 qMRI algorithm programed in Mathcad (Parametric Technology Corporation, Needham, MA) whereby the number of echoes used in the semi-logarithmic linear regression varies automatically from pixel to pixel depending on noise level. The T2 values of agarose gels have been measured during the entire gelation phase transition process at four different concentrations. The T2 versus time (temperature) curves of all the four concentrations shows a rapid drop at about 24 minutes (T~40°C) at which time the gelation phase transition begins. At all temperatures, T2 decreases as a function of increasing agarose concentration. The data shows similar behaviors for all concentrations with a phase transition characterized by a drastic drop in T2 occurring while the temperature drops by approximately 8°C. These results may be useful for testing theoretical models of the Nuclear Magnetic Resonance (NMR) T2 relaxation properties during tissue formation. Quantitative magnetic resonance imaging (qMRI) differs sharply from conventional directly acquired MRI in that objective measures [such as the trio of the basis MR properties: longitudinal relaxation (T1), T2 and Proton Density (PD)] are used for analysis as well as further post-processing rather than relative signal intensities. Q-MRI portrays the spatial distribution of absolute biophysical parameter measurements on a pixel-by-pixel basis; Kevin J. Chang et al 2005