Medical imaging is widely used in current healthcare and research settings for various purposes such as diagnosis, treatment options, patient monitoring, longitudinal studies, etc. The two most commonly used imaging modalities in the United States are Computed Tomography (CT) and Magnetic Resonance Imaging (MRI). Raw images acquired via CT or MRI need to undergo a variety of processing steps prior to being used for the purposes explained above. These processing steps include quality control, noise reduction, anatomical segmentation, tissue classification, etc. However, since medical images often include millions of voxels (smallest 3D units in the image containing information) it is extremely challenging to process them manually by relying on visual inspection and the experience of trained clinicians. In light of this, the field of medical imaging is seeking ways to automate data processing. With the impressive performance of Artificial Intelligence (AI) in the field of Computer Vision, researchers in the medical imaging community have shown increasing interest in utilizing this powerful tool to automate the task of processing medical imaging data. Despite AI’s significant contributions to the medical imaging field, large cohorts of data still remain without optimized and robust AI-based tools to process images efficiently and accurately. This thesis focuses on exploiting large cohorts of CT and MRI data to design AI-based methods for processing medical images using weakly-supervised and supervised learning strategies, as well as mathematical (and/or statistical) modeling and signal processing methods. In particular, we address four image processing problems in this thesis. Namely: 1) We propose a weakly-supervised deep learning method to automate binary quality control of diffusion MRI scans into ‘poor’ and ‘good’ quality classes; 2) We design a weakly-supervised deep learning framework to learn and detect visual patterns related to a set of different artifact categories considered in this work, in order to identify major artifact types present in dMRI volumes; 3) We develop a supervised deep learning method to classify multiple lung texture patterns with association to Emphysema disease on human lung CT scans; 4) We investigate and characterize the properties of two types of negative BOLD response elicited in human brain fMRI scans during visual stimulation using mathematical modeling and signal processing tools. Our results demonstrate that through the use of artificial intelligence and signal processing algorithms: 1) dMRI scans can be automatically categorized into two quality groups (i.e., ‘poor’ vs ‘good’) with a high classification accuracy, enabling rapid sifting of large cohorts of dMRI scans to be utilized in research or clinical settings; 2) Type of the major artifact present in ‘poor’ quality dMRI volumes can be identified robustly and automatically with high precision enabling exclusion/correction of corrupt volumes according to the artifact type contaminating them; 3) Multiple lung texture patterns related to Emphysema disease can be automatically and robustly classified across various large cohorts of CT scans enabling investigation of the disease through longitudinal studies on multiple cohorts; 4) Negative BOLD responses of different categories can be fully characterized on fMRI data collected from visual stimulation of human brain enabling researchers to better understand the human brain functionality through studying cohorts of fMRI scans.

Simultaneous EEG-fMRI is a multi-modal neuroimaging technique where hemodynamic activity across the brain is measured at millimeter spatial resolution using functional magnetic resonance imaging (fMRI) while electrical activity at the scalp is measured at millisecond resolution using electroencephalography (EEG). EEG-fMRI is significantly more challenging to collect than either modality on its own, due to electromagnetic coupling and interference between the modalities. The rationale for collecting the two modalities together is that, given the complementary spatial and temporal resolutions of the individual modalities, EEG-fMRI has the potential as a non-invasive neuroimaging technique for recovering the latent source space of neural activity. Inferring this latent source space and fusing these modalities has been a main challenge for realizing the potential of simultaneous EEG-fMRI for human neuroscience. In this thesis we develop a principled and interpretable approach to address this inference problem. We build on the knowledge of the generative processes underlying image/signal formation of each of the modalities, traditionally viewed via linear mappings, and recast this into a framework which we refer to as “neural transcoding”. The idea of neural transcoding is to generate a signal of one neuroimaging modality from another by first decoding it into a latent source space and then encoding it into the other measurement space. We implement this transcoding via deep architectures based on convolutional neural networks. We first develop a basic transcoding architecture and test it on simulated EEG-fMRI data. Evaluation on simulated data enables us to assess the model’s ability to recover a latent source space given known ground truth. We then extend this architecture and add a cycle consistency loss to create a cycle-CNN transcoder and show that it outperforms, in terms of the fidelity of recovered source space, both the basic transcoder as well as traditional source estimation techniques, even when we provide those techniques detailed information about the image generation process. We then assess the performance of the cycle-CNN transcoder on real simultaneous EEG-fMRI datasets including an auditory oddball dataset and a three-choice visual categorization dataset. Without any prior knowledge of either the hemodynamic response function or leadfield matrix, the transcoder is able to exploit the temporal and spatial relationships between the modalities and latent source spaces to learn these mappings. We show, for real EEG-fMRI data, the modalities can be transcoded from one to another, and that the transcoded results for unseen test data, have substantial correlation with the ground truth. In addition, we analyze the source space of the transcoders and observe latent neural dynamics that could not be observed with either modality alone--e.g., millimeter by millisecond dynamics of cortical regions representing motor activation and somatosensory feedback for finger movement. Collectively, this thesis demonstrates how one can incorporate a principled understanding of the generative process underlying biomedical image/signal formation in a deep learning framework to build models that are interpretable and symmetrically combine both modalities. It also potentially enables a new type of low-cost computational neuroimaging -- i.e., generating an ``expensive" fMRI BOLD image from ``low cost" EEG data.

No description available.

This work considered the frequency-modulated balanced steady-state free precession (fm-bSSFP) sequence as a tool to provide banding free bSSFP MR images. The sequence was implemented and successfully applied to suppress bandings in various in vitro and in vivo examples. In combination with a radial trajectory it is a promising alternative for standard bSSFP applications. First, two specialized applications were shown to establish the benefits of the acquisition strategy in itself. In real time cardiac imaging, it was shown that the continuous shift in frequency causes a movement of the bandings across the FOV. Thus, no anatomical region is constantly impaired, and a suitable timeframe can be found to examine all important structures. Furthermore, a combination of images with different artifact positions, similar to phase-cycled acquisitions is possible. In this way, fast, banding-free imaging of the moving heart was realized. Second, acquisitions with long TR were shown. While standard bSSFP suffers from increasing incidence of bandings with higher TR values, the frequency-modulated approach provided banding free images, regardless of the TR. A huge disadvantage of fm-bSSFP, in combination with the radial trajectory, is the decrease in signal intensity. In this work a specialized reconstruction method, the multifrequency reconstruction for frequency-modulated bSSFP (Muffm), was established, which successfully compensated that phenomena. The application of Muffm to several anatomical sites, such as inner ear, legs and cardiac acquisitions, proofed the advantageous SNR of the reconstruction. Furthermore, fm-bSSFP was applied to the clinically highly relevant task of water-fat separation. Former approaches of a phase-sensitive separation procedure in combination with standard bSSFP showed promising results but failed in cases of high inhomogeneity or high field strengths where banding artifacts become a major issue. The novel approach of using the fm-bSSFP acquisition strategy with the separation approach provided robust, reliable images of high quality. Again, losses in signal intensity could be regained by Muffm, as both approaches are completely compatible. Opposed to conventional banding suppression techniques, like frequency-scouts or phase-cycling, all reconstruction methods established in this work rely on a single radial acquisition, with scan times similar to standard bSSFP scans. No prolonged measurement times occur and patient time in the scanner is kept as short as possible, improving patient comfort, susceptibility to motion or physiological noise and cost of one scan. All in all, the frequency-modulated acquisition in combination with specializes reconstruction methods, leads to a completely new quality of images with short acquisition times. / In dieser Arbeit wird eine Modifikation der balanced steady-state free precession (bSSFP) Sequenz betrachtet. Die frequenzmodulierte bSSFP-Sequenz (fm-bSSFP) kann die sonst typischen Band-Artefakte in bSSFP-MR-Bildern verhindern. Die Sequenz wurde im Rahmen der Arbeit am MR-Scanner implementiert und erfolgreich in verschiedenen in-vitro- und in-vivo-Beispielen angewendet. In Kombination mit einer radialen Trajektorie erwies es sich als eine vielversprechende Alternative für alle Standard-bSSFP Anwendungen. Zuerst wurden zwei spezialisierte Anwendungen gezeigt, um die Vorteile der Akquisitionsstrategie an sich darzustellen. Am Beispiel der Echtzeit-Herzbildgebung konnte mit Hilfe der kontinuierlichen Frequenzverschiebung eine Bewegung der Bänder über das FOV erzeugt werden. Somit wird keine anatomische Region ständig von Artefakten überlagert und für jeden Bereich kann ein geeigneter Zeitrahmen gefunden werden, um die wichtigen Strukturen darzustellen und zu untersuchen. Darüber hinaus ist eine Kombination von Bildern mit verschiedenen Artefaktpositionen möglich, ähnlich zu mehreren Aufnahmen mit verschiedenen Phasenzyklen. Auf diese Weise wurde eine schnelle Bildgebung des sich bewegenden Herzens ohne Bandartefakte realisiert. Zusätzlich wurden Aufnahmen mit langen Repetitionszeiten (TR) untersucht. Während in der Standard-bSSFP die Häufigkeit von Bandartefakten mit steigendem TR-Wert zunimmt, lieferte der frequenzmodulierte Ansatz Banding-freie Bilder unabhängig vom TR. Ein großer Nachteil von fm-bSSFP in Kombination mit der radialen Trajektorie ist der Verlust von Signalintensität bei der Rekonstruktion. In dieser Arbeit wurde eine spezielle Rekonstruktionsmethode namens Muffm (mulitfrequency reconstruction for frequency-modulated bSSFP) etabliert, die diesen Verlust erfolgreich kompensieren kann. Die Anwendung von Muffm an verschiedenen anatomischen Strukturen, wie Innenohr, Bein und Herzaufnahmen, bestätigte das vorteilhafte Signal-zu-Rausch-Verhältnis, dass durch die spezielle Rekonstruktion gewonnen werden kann. Darüber hinaus wurde die fm-bSSFP auf die klinisch interessante Wasser-Fett-Trennung angewandt. Frühere Ansätze eines phasenempfindlichen Trennverfahrens in Kombination mit Standard-bSSFP zeigten vielversprechende Ergebnisse, scheiterten jedoch in Fällen hoher Inhomogenität oder hoher Feldstärken an den auftretenden Bandartefakten. Der neue Ansatz, diesen Separationsalgorithmus mit der fm-bSSFP-Akquisitionsstrategie zu verbinden, lieferte robuste, zuverlässige Bilder von hoher Qualität. Auch hier konnten entstehende Verluste in der Signalintensität durch Muffm zurückgewonnen werden, da beide Ansätze vollständig kompatibel sind. Im Gegensatz zu herkömmlichen Bandunterdrückungstechniken, wie Frequenz-Scouts oder die Aufnahme mehrerer Bilder mit verschiedenen Phasenzyklen, beruhen alle in dieser Arbeit etablierten Rekonstruktionsverfahren auf einer einzigen radialen Aufnahme. Die Messzeiten sind daher identisch zur Aufnahme einer Standard-bSSFP Messung. Das Verfahren ermöglicht eine deutliche Verkürzung der Aufenthaltsdauer im Scanner bei einer gleichzeitigen Garantie ein artefaktfreies Bild zu erhalten. Damit ist es insbesondere für Patienten von Vorteil, die unter Platzangst oder sonstigen Beschwerden leiden, die ein langes Stillliegen erschweren. Außerdem werden Bewegungsartefakte, physiologisches Rauschen und nicht zuletzt die Kosten eines Scans minimiert. Insgesamt bietet die frequenzmodulierte bSSFP Aufnahme in Kombination mit spezialisierten Rekonstruktionsverfahren neue Möglichkeiten zur schnellen Aufnahme von Bildern ohne Bandartefakte.

No description available.

Depression has increased among adolescents and adults over the last decade. Effective treatments and techniques to improve personal well being and disorders like depression are much needed. Mindfulness is a psychological technique that involves actively paying attention to one's awareness of the present moment with nonjudgmental acceptance (Kabat-Zinn, 1990). While the effectiveness of mindfulness interventions has been widely studied, relatively little research has been done on the effects of mindfulness interventions on depression using neuroimaging techniques such as functional magnetic resonance imaging (fMRI) or electroencephalography (EEG). This systematic review includes four studies that investigates the effectiveness of mindfulness interventions on depression measured by fMRI or EEG and different rating scales measuring depression. The results from this systematic review shows that mindfulness interventions may have an effect on depression. Two of the studies (Ferri et al., (2017); Yang et al., (2016), found significant differences in the rating scales for depression. The studies also suggest that mindfulness interventions can impact the brain regions involved in negative emotional processing in individuals with depression, such as the default mode network (DMN) and the dorsal anterior cingulate cortex (dACC).

Carotid atherosclerosis, one of the leading causes of ischemic stroke worldwide, can induce severe narrowing or even occlusion of the vessel, restricting blood flow to the brain and resulting in perfusion deficits. The plaque that has a high probability of undergoing rapid progression or future ruptures is defined as “vulnerable plaque”. Identifying vulnerable plaque is of great importance in clinical carotid atherosclerosis imaging. To date, a multi-contrast magnitude-based MR approach with blood suppression technique has been widely used to detect vulnerable plaque features. However, due to the limitations of magnitude-based methods, developing new MR techniques that have better sensitivity to hemorrhage and calcification is of great interest. Quantitative Susceptibility Mapping (QSM) is a technique that utilizes the MR phase information and has been widely used for quantifying the tissue susceptibility in the brain. The susceptibility contrast is extremely sensitive to hemorrhage and calcium which makes QSM a potential tool for carotid plaque imaging to identify intraplaque hemorrhage (IPH) and calcification. However, existing QSM methods have not been successfully implemented in the neck due to several challenges. The presence of air/tissue interface, plaque that has high susceptibility, and fat surrounding the carotid arteries can cause severe phase aliasing and other problems that will induce errors in the resultant susceptibility maps. To overcome these challenges and thus, develop a robust method for carotid QSM, a protocol that includes both data acquisition strategy and post-processing methods is proposed. For data acquisition, four echoes including two water/fat in-phase echoes and two water/fat out-of-phase echoes were collected. For data post-processing, temporal domain algorithm Catalytic Multiecho Phase Unwrapping Scheme (CAMPUS) was used to unwrap the phase images and local QSM was proposed. This protocol is able to properly unwrap the phase images even with the presence of high susceptibility plaque and eliminate the water/fat chemical shift effect in QSM reconstructions which will generate reliable susceptibility maps. From our results, the proposed QSM protocol has demonstrated the ability to generate reliable susceptibility maps and excellent sensitivity to IPH and calcification. Combining QSM with existing magnitude-based methods will lead to a major improvement in the diagnosis of carotid atherosclerosis. / Thesis / Master of Applied Science (MASc)

BACKGROUND: Autosomal dominant polycystic kidney disease (ADPKD) is characterized by gradual kidney enlargement and cyst growth prior to loss of kidney function. The Consortium for Radiologic Imaging Studies in Polycystic Kidney Disease (CRISP) created a standard magnetic resonance imaging (MRI) protocol to be used for ADPKD patients to determine if changes in total kidney volumes can be detected over a short period of time, and if they correlate with decline in renal function early in the disease course. CRISP guided researchers and physicians to use a T1-weighted sequence with gadolinium contrast to measure kidney volumes. After the Food and Drug Administration discouraged the use of gadolinium contrast in individuals with kidney diseases, total kidney volume measured by MRI for ADPKD patients was done using the T1-weighted pulse sequence without contrast enhancement. STUDY OBJECTIVE: The retrospective cohort study will aim to assess reliability of the T2 sequence and total kidney volume measurements compared to total kidney volume measurements performed on a T1 sequence. METHODS: The study collected intra-reader and inter-reader cases from four imaging studies, each with an abdominal MRI performed. Repeated volume measurements were performed within an individual reader (intra-reader) and between different readers (inter-readers). The stereology method was used to quantify kidney volume from T1 images for three studies and T2 images for one study. Mean and standard deviation were used to analyze volume differences between repeated measurements for intra-reader and inter-reader data for each MRI sequence. The intra-class correlation coefficient and Bland-Altman plot were used to describe correlation between kidney volumes, for intra-reader and inter-reader data respectively. RESULTS: Analyses show a significant difference in the repeated volume measurements from the T1 sequence in inter-reader data. Reliability for the T2 and T1 sequence was represented by high correlations in both the intra-reader and inter-reader total kidney volumes. CONCLUSION: MR measures of total kidney volume are reliable in patients when measured on both the T1 sequence and the T2 sequence. ADPKD kidney volumes for future clinical trials can be reliably measured on either sequence.

The possibility to perform virtual, non-invasive, quantitative, in vivo histological assessments might revolutionize entire fields, among which clinical and cognitive neurosciences. Magnetic Resonance Imaging (MRI) is an ideal non-invasive imaging technique to achieve these goals. Tremendous advancements in the last decades have favored the transition of MRI scanners from “imaging devices” to “measurement devices” (Novikov, 2021), thus capable to yield measurements in physical units, which might be further combined to provide quantities describing histological properties of substrates. A central role in this community endeavor has been played by diffusion-weighted MRI (dMRI), which by measuring the dynamics of spin diffusion, allows inferences on geometrical properties of tissues. Yet, conventional dMRI methodologies suffer from poor specificity. In this thesis, techniques aiming at improving the specificity of microstructural descriptions have been explored in dMRI datasets supporting an increasing level of complexity of the dMRI signal representations. Applications in individuals with different age range, in different populations, and for different MRI scanner fields, have been considered. Firstly, tractography has been combined with Diffusion Tensor Imaging (DTI), an along-tract framework, and morphometry, in the study of the microstructure of the optic radiations in different groups of blind individuals. Secondly, DTI has been combined with Free-Water Imaging (FWI) to monitor the effect of proton-irradiation in a pediatric brain tumor case study. Thirdly, FWI and Diffusion Kurtosis Imaging (DKI) have been combined with an advanced thalamic segmentation framework to study the associations between motor performance and thalamic microstructure in a cohort of individuals affected by Parkinson’s disease. Finally, the largest contribution of this thesis is represented by the adaptation of the Correlation Tensor Imaging - a technique increasing the specificity of DKI harnessing Double Diffusion Encoding previously applied only in preclinical settings - for a clinical 3 T scanner. The ensuing investigation revealed new important insights on the sources of diffusional kurtosis, in particular of the microscopic kurtosis (μK), a component so far neglected by contemporary neuroimaging techniques, which might carry an important clinical role (Alves et al., 2022), and can now be accessed by clinical scanners. In conclusion, strategies to increase the specificity of microstructural descriptions in the brain are presented for different datasets, and their strength and limitations are discussed.

No description available.