Deep learning has overwhelmingly impacted post-acquisition image-processing tasks, however, there is increasing interest in more tightly coupled computational imaging approaches, where models, computation, and physical sensing are intertwined. This dissertation focuses on how to leverage the expressive power of deep learning in image reconstruction. We use deep learning in both the sensor data domain and the image domain to develop new fast and efficient algorithms to achieve superior quality imagery. Metal artifacts are ubiquitous in both security and medical applications. They can greatly limit subsequent object delineation and information extraction from the images, restricting their diagnostic value. This problem is particularly acute in the security domain, where there is great heterogeneity in the objects that can appear in a scene, highly accurate decisions must be made quickly, and the processing time is highly constrained. Motivated primarily by security applications, we present a new deep-learning-based MAR approach that tackles the problem in the sensor data domain. We treat the observed data corresponding to dense, metal objects as missing data and train an adversarial deep network to complete the missing data directly in the projection domain. The subsequent complete projection data is then used with an efficient conventional image reconstruction algorithm to reconstruct an image intended to be free of artifacts. Conventional image reconstruction algorithms assume that high-quality data is present on a dense and regular grid. Using conventional methods when these requirements are not met produces images filled with artifacts that are difficult to interpret. In this context, we develop data-domain deep learning methods that attempt to enhance the observed data to better meet the assumptions underlying the fast conventional analytical reconstruction methods. By focusing learning in the data domain in this way and coupling the result with existing conventional reconstruction methods, high-quality imaging can be achieved in a fast and efficient manner. We demonstrate results on four different problems: i) low-dose CT, ii) sparse-view CT, iii) limited-angle CT, and iv) accelerated MRI. Image domain prior models have been shown to improve the quality of reconstructed images, especially when data are limited. A novel principled approach is presented allowing the unified integration of both data and image domain priors for improved image reconstruction. The consensus equilibrium framework is extended to integrate physical sensor models, data models, and image models. In order to achieve this integration, the conventional image variables used in consensus equilibrium are augmented with variables representing data domain quantities. The overall result produces combined estimates of both the data and the reconstructed image that is consistent with the physical models and prior models being utilized. The prior models used in both image and data domains in this work are created using deep neural networks. The superior quality allowed by incorporating both data and image domain prior models is demonstrated for two applications: limited-angle CT and accelerated MRI. A major question that arises in the use of neural networks and in particular deep networks is their stability. That is, if the examples seen in the application environment differ from the training environment will the performance be robust. We perform an empirical stability analysis of data and image domain deep learning methods developed for limited-angle CT reconstruction. We consider three types of perturbations to test stability: adversarially optimized, random, and structural perturbations. Our empirical analysis reveals that the data-domain learning approach proposed in this dissertation is less susceptible to perturbations as compared to the image-domain post-processing approach. This is a very encouraging result and strongly supports the main argument of this dissertation that there is value in using data-domain learning and it should be a part of our computational imaging toolkit.

Ancestry estimation of skeletonized remains by forensic anthropologists is conducted through comparative means, and a lack of population-specific data results in possible misclassifications. This is especially germane to individuals of Latin American ancestry. Generally, each country in Latin America can trace their ancestry to three parental groups: Indigenous, European, and African. However, grouping all Latin American individuals together under the broad “Hispanic” category ignores the specific genetic contributions from each parental group, which is variable and dependent on the population histories and sociocultural dynamics of each country. This study analyzes the craniometric ancestry of Hispaniola (the Dominican Republic and Haiti) using the island’s history, along with 190 cranial Computed Tomography (CT) scans (f = 103; m = 87), to determine similarities and differences between the two groups. A total of 12 linear discriminant function analyses produced cross-validated classification accuracies of 75.0 – 83.3% for females, 71.8 – 87.5% for males, and 72.0 – 82.2% for pooled sexes. This study demonstrates that, despite sharing a small island, Dominican and Haitian individuals can be differentiated with a fair amount of statistical certainty, which is possible due to complex socio-cultural, -political, and –demographic factors that have maintained genetic heterogeneity. Moreover, the discriminant functions provided here can be used by the international forensic science community to identify individuals living on Hispaniola.

Bakgrund: Stenar i njurar respektive urinvägar är ett växande problem både i Sverige samt i hela världen. Anledningen till den ökade incidensen av njursten anses vara oklar, däremot kan ändrade kostvanor förslagsvis vara bidragande orsak. Datortomografi (DT) anses vara förstahandsalternativet vid utredning av misstänkta njurstenar där DT har bra förmåga att kartlägga antalet stenar, bestämma storlek samt lokalisera stenarna. Tidigare forskning visar att ultra-lågdos samt lågdos har höga färdigheter i sensitivitet samt specificitet. Däremot avkastas jämförelser av detektering av stenar med standarddos DT. Syfte: Syftet med denna studie var att utforska om hur exponeringen skiljer sig med användandet av låg-dos protokoll i jämförelse med standardprotokoll. Utforska om hur mycket stråldosen kunde sänkas med lågdosprotokoll samt förmågan att detektera stenar. Fokus kommer att ligga på stråldosreducering samt specificitet och sensitivitet i utredning för njurstensmisstanke med datortomografi. Metod: Denna litteraturstudie använder sig av databaser som Pubmed för eftersökningen av vetenskapliga artiklar. Materialet samlades in och bearbetades i enlighet med Forsberg & Wengströms riktlinjer. Resultat: Resultatet till denna studie baseras på 10 valda artiklar från databasen Pubmed. Resultatet presenteras i tre underkategorier, storlek av sten, exponering samt sensitivitet/specificitet. Konklusion: Lågdosprotokoll kunde konkurrera med standardprotokoll gällande detektion av stenar, stenar under 3 mm kan vara svåra att detektera med lågdosprotokoll. Stråldosen reducerades med mer än hälften och bibehöll samtidigt tillräckligt bra kvalitet. Patienter bestrålas med mer än dubbelt så mycket i effektiv dos (ED) med användning av standardprotokoll än med låg-dos protokoll.

This study is carried out to develop the workflow from image acquisition to numerical simulation for asphalt concrete (AC) microstructure. High resolution computed tomography (CT) images are acquired and the image quality is improved using digital image processing (DIP). Non-uniform illumination which results in inaccurate phase segmentation is corrected by applying an illumination profile to correct the background and flat-fields in the image. Distance map based watershed segmentation is used to accurately segment the phases and separate the aggregates. Quantitative analysis of the microstructure is used to determine the phase volumetric relationship and aggregates characteristics. The results of the phase reconstruction and internal structure quantification using this procedure shows a very high level of reliability. Numerical simulations are carried out in Two dimensions (2D) and Three dimensions (3D) on the processed AC microstructure. Finite element analysis (FEM) is used to capture the strength and deformation mechanisms of the AC microstructure. The micromechanical behaviour of the AC is investigated when it is considered as a continuum and when considered as a multi-phase model. The results show that the size and arrangement of aggregates determines the stress distribution pattern in the mix.

The behavior of the asphalt mixtures under large deformations, for example an unconfined compaction is of high practical importance. Quantitative measurement of the spatial distribution of internal structure of asphalt mixtures is crucial to study deformation behavior of asphalt mixtures. Deformation of granular material under an unconfined compaction is investigated in this study, as a groundwork for further research on deformation behavior of asphalt mixtures. Two sets of 3D images of specimens are obtained using X-Ray computed tomography (CT) under an unconfined compaction. Digital image analysis procedure is developed to segment different phases for characterizing spatial distribution of internal structure. Comparative volumetric relationship before and after compaction showed that air void distribution is not changed heavily due to absence of interlocking. Initial and final spatial positions of individual granules are investigated to trace their movement under compaction. It is shown that X-Ray CT could be a useful tool to characterize internal structure of asphalt mixtures and its evolution during deformation.

This master thesis evaluated the performance of a micro-CT system consisting of Hamamatsu microfocus X-ray source L10951-04 and CMOS flat panel C7942CA-22. The X-ray source and flat panel have been characterised in terms of dark current, image noise and beam profile. Additionally, the micro-CT system’s spatial resolution, detector lag and detector X-ray response have been measured. Guidance for full image correction and methods for characterisation and performance test of the X-ray source and detector is presented. A spatial resolution of 7 lp/mm at 10 % MTF was measured. A detector lag of 0.3 % was observed after ten minutes of radiation exposure. The performance of the micro-CT system was found to be sufficient for high resolution X-ray imaging. However, the detector lag effect is strong enough to reduce image quality during subsequent image acquisition and must either be avoided or corrected for.

The complexity of components fabricated in today's industry is ever increasing. This increase is partly due to market pressure but it is also a result from progress in fabrication technologies that opens up new possibilities. The increased use of additive manufacturing and multi-material systems, especially, has driven the complexity of parts to new heights. The new complex material systems brings benefits in many areas such as; mechanical properties, weight optimisation, and sustainability. However, the increased complexity also makes material integrity investigations and dimensional control more difficult. In additive manufacturing, for example, internal features can be fabricated which cannot be seen or measured with conventional tools. There is thus a need for non-destructive inspection methods that can measure these geometries. Such a method is X-ray computed tomography. Computed tomography utilizes the X-rays ability to penetrate material to create 3D digital volumes of components. Measurements and material investigations can be performed in these volumes without any damage to the investigated component. However, computed tomography in material science is still not a fully mature method and there are many uncertainties associated with the investigation technique. In the work presented in this thesis geometries fabricated by various additive manufacturing processes have been investigated using computed tomography. Also in this work, a dual-energy computed tomography tool has been developed with the aim to increase the measurement consistency of computed tomography when investigating complex geometries and material combinations. / MultiMatCT

Additive manufacturing (AM) allows building complex shapes with high accuracy. The X-ray Computed Tomography (XCT) is one of the promising non-destructive evaluation techniques for the evaluation of subsurface defects in an additively manufactured component. Automatic defect detection and segmentation methods can assist part inspection for quality control. However, automatic detection and segmentation of defects in XCT data of AM possess challenges due to contrast, size, and appearance of defects. In this research different deep learning techniques have been applied on publicly available XCT image datasets of additively manufactured cobalt chrome samples produced by the National Institute of Standards and Technology (NIST). To assist the data labeling image processing techniques were applied which are median filtering, auto local thresholding using Bernsen’s algorithm, and contour detection. A convolutional neural network (CNN) based state-of-art object algorithm YOLOv5 was applied for defect detection. Defect segmentation in XCT slices was successfully achieved applying U-Net, a CNN-based network originally developed for biomedical image segmentation. Three different variants of YOLOv5 which are YOLOv5s, YOLOv5m, and YOLOV5l were implemented in this study. YOLOv5s achieved defect detection mean average precision (mAP) of 88.45 % at an intersection over union (IoU) threshold of 0.5. And mAP of 57.78% at IoU threshold 0.5 to 0.95 using YOLOv5M was achieved. Additionally, defect detection recall of 87.65% was achieved using YOLOv5s, whereas a precision of 71.61 % was found using YOLOv5l. YOLOv5 and U-Net show promising results for defect detection and segmentation respectively. Thus, it is found that deep learning techniques can improve the automatic defect detection and segmentation in XCT data of AM.

Sarcoidosis is a multisystem granulomatous disease commonly involving the lungs and mediastinal lymph nodes with the exact etiology being unclear. The simultaneous presence of malignant disease such as breast cancer and sarcoidosis has been reported. Sarcoidosis preceding a diagnosis of malignancy and that occurring years after treatment of malignant disease has been noted in the past. The presence of sarcoidosis in the setting of malignant disease carries a high risk of misdiagnosis. In this article, we report the case of a 45-year-old female with stage IA invasive ductal carcinoma of left breast that was in remission for 2 years; however, radiological imaging including magnetic resonance imaging of thoracic spine and positron emission tomography–computed tomography scanning were highly suspicious for malignant disease metastasis versus lymphoma with the widespread lymphadenopathy. Multiple tissue biopsies with histopathological evaluation allowed us to definitively exclude malignant disease metastasis and to correctly diagnose her atypical presentation of sarcoidosis.

Context: Scimitar syndrome is a congenital anomaly of pulmonary venous return where right pulmonary artery drains into right side other heart, instead of the left side, causing pulmonary hypertension resulting in shortness of breath, recurrent lower respiratory tract infections, chest pain, and fatigue. Early diagnosis and surgical intervention would correct this congenital anomaly reducing morbidity and complications in otherwise healthy young patients. Case Report: We present a case of an 18-year-old female who presented with exertional shortness of breath, fatigue, and recurrent lower respiratory tract infections. She had unremarkable physical examination but chest x-ray showed an abnormal opacity next to right heart border. Computed tomography (CT) chest was performed that showed possible scimitar syndrome. Transesophageal echocardiogram (TEE) and right heart catheterization (RHC) confirmed the diagnosis. Conclusion: Scimitar syndrome is a very rare congenital anomaly of pulmonary venous return. It is usually diagnosed in early childhood but the diagnosis may be delayed until later in adulthood. The consequences are pulmonary hypertension, right-sided heart failure, and frequent pulmonary infections resulting in increased morbidity, mortality, and frequent doctor visits for otherwise healthy young patients.