No description available.

Recent advancements in empirical and certified robustness have shown promising results in developing reliable and deployable Deep Neural Networks (DNNs). However, most evaluations of DNN robustness have focused on testing models on images from the same distribution they were trained on. In real-world scenarios, DNNs may encounter dynamic environments with significant distribution shifts. This thesis aims to investigate the interplay between empirical and certified adversarial robustness and domain generalization. We take the first step by training robust models on multiple domains and evaluating their accuracy and robustness on an unseen domain. Our findings reveal that: (1) both empirical and certified robustness exhibit generalization to unseen domains, and (2) the level of generalizability does not correlate strongly with the visual similarity of inputs, as measured by the Fréchet Inception Distance (FID) between source and target domains. Furthermore, we extend our study to a real-world medical application, where we demonstrate that adversarial augmentation significantly enhances robustness generalization while minimally affecting accuracy on clean data. This research sheds light on the importance of evaluating DNNs under real-world distribution shifts and highlights the potential of adversarial augmentation in improving robustness in practical applications.

No description available.

No description available.

Intra-cellular sodium (23Na) concentration is directly related to cellular health. Thus, sodium magnetic resonance imaging (MRI) can provide metabolic information on tissue health that a routine clinical (proton) MRI cannot. 23Na-MRI could be a valuable tool to assist physicians in the diagnosis, prognosis, and monitoring of a variety of pathologies. However, due to factors that include quantum mechanical limitations and biological restrictions, the signal-to-noise ratio (SNR) of a sodium scan is much lower than that of a standard proton scan, which limits the practicality of 23Na-MRI in a clinical setting. This project looks to improve the viability of 23Na-MRI and focuses on an often overlooked facet of MRI development, the radio frequency (RF) coil. Fractal antennas have been used in telecommunication systems for years, and are generally exploited for their compact nature, allowing for the same performance of a larger antenna, in a smaller space. They have also been shown to be capable of a wider transmission bandwidth (BW) than a standard antenna and with MRI applications they have been shown to provide a small SNR increase in proton imaging. It is hypothesized that a surface coil with a Koch snowflake fractal geometry can provide increased SNR for a sodium MRI scan, compared to that of a standard circular geometry coil, by producing a more homogeneous magnetic field in both space and frequency. To test the hypothesis two coils, one circular and the other a Koch snowflake fractal, were simulated. The simulated magnetic fields were compared on their homogeneity and magnitude before the two coils were constructed and implemented with a variety of sodium MRI scans. B1+ maps were acquired to measure RF field homogeneity, and SNR was determined for both coil geometries. The coils were also tested for their homogeneity over varied transmit BWs by comparing images with various field of view (FOV) sizes. Finally the coils were compared for clinical viability in a test of healthy human knee imaging. The circular coil had a more homogeneous B1+ field than the fractal at depths between 10-40mm, and had a higher SNR in its produced images. The circular coil acquired more signal in vivo which provided a higher detail image, but the fractal coil's SNR was higher due to reduced noise. The fractal coil performed better over a wider BW which indicates that further research should be conducted into the applications of fractal coils in multi-nuclear MRI scans. / Thesis / Master of Applied Science (MASc)

No description available.

Modern clinical diagnoses and treatments have been increasingly reliant on medical imaging techniques. In return, medical images are required to provide more accurate and detailed information than ever. Aside from the evolution of hardware and software, multimodal imaging techniques offer a promising solution to produce higher quality images by fusing medical images from different modalities. This strategy utilizes more structural and/or functional image information, thereby allowing clinical results to be more comprehensive and better interpreted. Since their inception, multimodal imaging techniques have received a great deal of attention for achieving enhanced imaging performance. In this work, a novel joint reconstruction framework using sparse computed tomography (CT) and magnetic resonance imaging (MRI) data is developed and evaluated. The method proposed in this study is part of the planned joint CT-MRI system which assembles CT and MRI subsystems into a single entity. The CT and MRI images are synchronously acquired and registered from the hybrid CT-MRI platform. However, since their image data are highly undersampled, analytical methods, such as filtered backprojection, are unable to generate images of sufficient quality. To overcome this drawback, we resort to compressed sensing techniques, which employ sparse priors that result from an application of L₁-norm minimization. To utilize multimodal information, a projection distance is introduced and is tuned to tailor the texture and pattern of final images. Specifically CT and MRI images are alternately reconstructed using the updated multimodal results that are calculated at the latest step of the iterative optimization algorithm. This method exploits the structural similarities shared by the CT and MRI images to achieve better reconstruction quality. The improved performance of the proposed approach is demonstrated using a pair of undersampled CT-MRI body images and a pair of undersampled CT-MRI head images. These images are tested using joint reconstruction, analytical reconstruction, and independent reconstruction without using multimodal imaging information. Results show that the proposed method improves about 5dB in signal-to-noise ratio (SNR) and nearly 10% in structural similarity measurements compared to independent reconstruction methods. It offers a similar quality as fully sampled analytical reconstruction, yet requires as few as 25 projections for CT and a 30% sampling rate for MRI. It is concluded that structural similarities and correlations residing in images from different modalities are useful to mutually promote the quality of image reconstruction. / Ph. D. / Medical imaging techniques play a central role in modern clinical diagnoses and treatments. Consequently, there is a constant demand to increase the overall quality of medical images. Since their inception, multimodal imaging techniques have received a great deal of attention for achieving enhanced imaging performance. Multimodal imaging techniques can provide more detailed diagnostic information by fusing medical images from different imaging modalities, thereby allowing clinical results to be more comprehensive to improve clinical interpretation. A new form of multimodal imaging technique, which combines the imaging procedures of computed tomography (CT) and magnetic resonance imaging (MRI), is known as the “omnitomography.” Both computed tomography and magnetic resonance imaging are the most commonly used medical imaging techniques today and their intrinsic properties are complementary. For example, computed tomography performs well for bones whereas the magnetic resonance imaging excels at contrasting soft tissues. Therefore, a multimodal imaging system built upon the fusion of these two modalities can potentially bring much more information to improve clinical diagnoses. However, the planned omni-tomography systems face enormous challenges, such as the limited ability to perform image reconstruction due to mechanical and hardware restrictions that result in significant undersampling of the raw data. Image reconstruction is a procedure required by both computed tomography and magnetic resonance imaging to convert raw data into final images. A general condition required to produce a decent quality of an image is that the number of samples of raw data must be sufficient and abundant. Therefore, undersampling on the omni-tomography system can cause significant degradation of the image quality or artifacts after image reconstruction. To overcome this drawback, we resort to compressed sensing techniques, which exploit the sparsity of the medical images, to perform iterative based image reconstruction for both computed tomography and magnetic resonance imaging. The sparsity of the images is found by applying sparse transform such as discrete gradient transform or wavelet transform in the image domain. With the sparsity and undersampled raw data, an iterative algorithm can largely compensate for the data inadequacy problem and it can reconstruct the final images from the undersampled raw data with minimal loss of quality. In addition, a novel “projection distance” is created to perform a joint reconstruction which further promotes the quality of the reconstructed images. Specifically, the projection distance exploits the structural similarities shared between the image of computed tomography and magnetic resonance imaging such that the insufficiency of raw data caused by undersampling is further accounted for. The improved performance of the proposed approach is demonstrated using a pair of undersampled body images and a pair of undersampled head images, each of which consists of an image of computed tomography and its magnetic resonance imaging counterpart. These images are tested using the proposed joint reconstruction method in this work, the conventional reconstructions such as filtered backprojection and Fourier transform, and reconstruction strategy without using multimodal imaging information (independent reconstruction). The results from this work show that the proposed method addressed these challenges by significantly improving the image quality from highly undersampled raw data. In particular, it improves about 5dB in signal-to-noise ratio and nearly 10% in structural similarity measurements compared to other methods. It achieves similar image quality by using less than 5% of the X-ray dose for computed tomography and 30% sampling rate for magnetic resonance imaging. It is concluded that, by using compressed sensing techniques and exploiting structural similarities, the planned joint computed tomography and magnetic resonance imaging system can perform imaging outstanding tasks with highly undersampled raw data.

This dissertation explores the application of self-supervised learning methods in dental radiology to address the challenges posed by limited data availability for training deep learning models. The overarching goal is to enhance the efficiency and accuracy of automated systems for various dental diagnostic tasks, including teeth numbering, detection of dental restorations, orthodontic appliances, implant systems, marginal bone level, and dental caries from panoramic radiographs, CBCT images, intra-oral 3D scans, and dental radiographs. Key contributions include the development of several novel approaches: Self-supervised Learning for Dental Panoramic Radiographs: Utilizing SimMIM and UM-MAE with Swin Transformer, we achieved significant improvements in teeth detection and instance segmentation, increasing the average precision by 13.4% and 12.8%, respectively, over baseline methods. Self-Distillation Enhanced Self-supervised Learning (SD-SimMIM): Enhancing SimMIM with self-distillation loss, we improved performance on teeth numbering, dental restoration detection, and orthodontic appliance detection tasks, demonstrating superior outcomes compared to other methods. DentalMAE for Intra-oral 3D Scans: Extending the mesh masked autoencoder (MeshMAE), DentalMAE evaluates predicted deep embeddings of masked mesh triangles, yielding better generalization and higher accuracy in teeth segmentation tasks. DEMAE for Dental CBCT Images: Proposing the Deep Embedding MAE (DEMAE), which measures the closeness of predicted deep embeddings of masked patches to their originals, we achieved significant accuracy improvements in teeth segmentation from CBCT images. Masked Deep Embedding (MDE) for Implant Detection: By leveraging MIM, we developed MDE to enhance dental implant detection, creating a comprehensive Implant Design Dataset (IDD) with expert annotations, significantly boosting detection performance. Deep Embedding of Patches (DEP) for Bone Loss Assessment: An extension of MAE, DEP improved the accuracy of marginal bone level detection, supported by the creation of a Bone Loss Assessment Dataset (BLAD) with detailed annotations. Masked Deep Embedding of Patches (MDEP) for Caries Detection: This method enhanced dental caries detection performance, validated on the CariesXrays dataset, demonstrating higher precision and recall rates compared to traditional baselines. Through these innovations, the dissertation establishes the efficacy of self-supervised learning in overcoming data scarcity in dental imaging, offering promising AI-driven solutions for improved diagnostics and patient care in dentistry. / Computer and Information Science

Yes / This article discusses the current evidence for the short- and long-term effects of concussion in sport and how occurrences of concussion should be managed. The article also considers the potential role of medical imaging in terms of assessing both acute and chronic head injuries. Greater awareness of when medical imaging could be used will aid the practitioner's understanding of its potential contribution while still maintaining the fundamental importance of clinical judgement.

Yes / Focus on patient experience and patient centred approaches within healthcare has substantially increased since the Picker Institute (a not for profit organisation) was established in the 1980′s [ [15] ]. The Picker Institute's vision for ‘the highest quality person centred care for all, always’ outlines eight principles of person-centred care which health care providers should strive for [ [15] ]: (1) Fast access to reliable healthcare advice [15]. (2) Effective treatment delivered by trusted professionals [15]. (3) Continuity of care and smooth transitions. [15] (4) Involvement and support for family and carers [15]. (5) Clear information, communication and support for self-care [15]. (6) Involvement in decisions and respect for preferences [15]. (7) Emotional support, empathy and respect [15]. (8) Attention to physical and environmental needs [15].