University of Macau / Faculty of Health Sciences

This thesis aimed to investigate the feasibility of two diffusion imaging techniques, Diffusion Tensor Imaging (DTI) and Diffusion Kurtosis Imaging (DKI), on detecting subtle physiological or pathological microstructural changes in normal-appearing neural tissues of human central nervous system. 　　　At first, ten patients with neuromyelitis optica (NMO) and twelve age- and gender-matched healthy subjects were recruited. DTI-derived indices including fractional anisotropy (FA), mean diffusivity (MD), axial and radial diffusivities were quantified in the lateral and dorsal columns of cervical spinal cord. Based on the regions of interest (ROIs) measurement, NMO patients showed reduced FA, increased MD and radial diffusivity compared to control subjects, while axial diffusivity did not show any significant difference. The three former DTI metrics also showed significant correlations with disability scores, and especially FA was found to be sensitive to mild NMO. Our results show that DTI-derived indices can quantitatively assess the white matter (WM) abnormalities with seemingly normal appearance in conventional MRI, and are associated with the level of clinical disability, suggesting that DTI may have great potential as a useful diagnostic tool in the clinical setting. 　　　DKI is an extension of conventional DTI to probe the non-Gaussian diffusion property in biological tissues. Besides the four conventional DTI-derived metrics, DKI also provide three additional kurtosis metrics (mean kurtosis (MK), axial and radial kurtosis). In the second study, ROI-based analysis was used to characterize age-related microstructural changes in WM, cortical and subcortical gray matter (GM) of 27 healthy adults (21~59 yrs). Though the volumes of GM and WM were still preserved, DTI-derived metrics can detect the subtle changes in WM and GM. Meanwhile, MK and radial kurtosis significantly increased in both caudate nucleus and putamen while Thalamus showed little aging effect in the diffusivity and kurtosis metrics but significantly decreased only in FA. Our results demonstrated that DKI is sensitive to detect the age-related alterations in neural microstructures at the stage of early aging. 　　　In addition, DKI has been applied to detect the pathological changes in the normal-appearing neural tissues of 18 patients with multiple sclerosis (MS), compared to 22 healthy controls. Diffuse WM abnormalities have been observed extensively in the brain, revealed by DKI-derived metrics. Though the volumetric and voxel-wise analysis revealed no significant changes in the volume of cortical GM, decreased FA and kurtoses with increased diffusivities in MS group were sensitive to disclose the subtle alterations in global and regional cortical GM tissues. Significant correlations have been found between FA in the global, frontal and temporal cortical GM in relapsing-remitting MS patients and their disability scores, suggesting FA as an important biomarker to monitor the disease progress in cortical GM. Moreover, elevated kurtosis indices in MS patients did not correlate with diffusivities in caudate nucleus, putamen and thalamus, suggesting these metrics may be vulnerable to different pathologic aspects of the disease. 　　　In conclusion, DKI is sensitive to neural alterations during normal aging and in MS pathologies, and can provide complementary information to conventional MRI and DTI. / published_or_final_version / Diagnostic Radiology / Doctoral / Doctor of Philosophy

Despite its potential, 66% of the world's population lacks access to magnetic resonance imaging (MRI). The main factors contributing to the uneven distribution of this imaging modality worldwide are the elevated cost and intricate nature of MRI systems coupled with the high level of knowledge and expertise required for its operation and maintenance. To improve its worldwide accessibility, MRI technology and techniques must undergo modifications to deliver a more cost-effective system that is easier to site and use without compromising on the diagnostic quality of the images. This thesis presents two deep learning methods, ArtifactID and GDCNet, developed for artifact detection and correction and tailored for their integration into accessible MRI systems. ArtifactID is targeted to resource-constrained settings where skilled personnel are scarce. It automates part of the quality assessment step, critical during image acquisition to ensure data quality and the success of downstream analysis or interpretation. This study utilized two types of T1-weighted neuroimaging datasets: publicly available and prospective. Combining the two, ArtifactID successfully identified wrap-around and rigid head motion in multi-field strength and multi-vendor data. We leveraged the public datasets for artifact simulation, model training, and testing. In contrast, prospective datasets were reserved for validation and testing and to assess the models’ performance in data representative of clinical and deployment settings. We trained individual convolutional neural networks for each artifact. The wrap-around models perform binary classification, while the multi-class motion classification model allows distinction between moderate and severe motion artifacts. Our models demonstrated strong agreement with ground truth labels and motion metrics and proved potential for generalization to various data distributions. Furthermore, Grad-CAM heatmaps allowed early identification of failure modes, artifact localization within the image, and fine-tuning the pre-processing steps. GDCNet correction applies to imaging techniques highly susceptible to local B0 deviations and systems whose design entails high B0 inhomogeneity. The method estimates a geometric distortion map by non-linear registration to a reference image. The self-supervised model, consisting of a U-Net and a spatial transform function unit, learned the correction by optimizing the similarity between the distorted and the reference images. We initially developed the tool for distortion correction of echo-planar imaging functional MRI images at 3 T. This method allows dynamic correction of the functional data as a distortion map is estimated for each temporal frame. For this model, we leveraged T1-weighted anatomical images as target images. We trained the model on publicly available datasets and tested it on in-distribution and out-of-distribution datasets consisting of other public datasets unseen during training and a prospectively acquired dataset. Comparing GDCNet to state-of-the-art EPI geometric distortion methods, our technique demonstrated statistically significant improvements in normalized mutual information between the corrected and reference images and 14 times faster processing times without requiring the acquisition of additional sequences for field map estimation. We adapted the GDCNet method for distortion correction of low-bandwidth images acquired in a 47 mT permanent magnet system. These systems are characterized by large B0 spatial inhomogeneity and low signal sensitivity. In this case, the model used high-field images or images acquired with higher acquisition bandwidths as reference. The goal was to exploit the signal-to-noise ratio improvements that low bandwidth acquisition offers while limiting geometric distortion artifacts in the images. We investigated two versions of the model using different similarity loss functions. Both models were trained and tested on an in vitro dataset of image-quality phantoms. Additionally, we evaluated the models’ generalization ability to an in vivo dataset. The models successfully reduced distortions to levels comparable to those of the high bandwidth images in vitro and improved geometric accuracy in vivo. Furthermore, the method indicated robust performance on reference images with large levels of noise. Incorporating the methods presented in this thesis into the software of a clinical MRI system will alleviate some of the barriers currently restricting the democratization of MR technology. First, automating the time-consuming process of artifact identification during image quality assessment will improve scan efficiency and augment expertise on-site by assisting non-skilled personnel. Second, efficient off-resonance correction during image reconstruction will ease the tight B0 homogeneity requirements of magnet design, allowing more compact and lightweight systems that are easier to refrigerate and site.

How does the brain integrate information from individual neurons? One efficient way to investigate systematic neuroscience is to record the whole brain down to singular neuron level. Caenorhabditis elegans, a 1 mm long, transparent nematode species, is ideally suited as a starting point. Every C. elegans hermaphrodite has a fixed set of 302 neurons. All neuron connections have been fully characterized by electron microscopy. Despite its small and simple nervous system, C. elegans exhibits a wide range of behaviors ranging from foraging, sleep to sexual activity. Recently, Yemini et al. genetically engineered a C. elegans strain where each neuron can be uniquely identified by its color code. This greatly facilitates comparison of neural recordings with literature as well as underlying connectomics. However, it is a daunting task to record the whole nervous system at cellular resolution of a freely moving worm. The imaging system needs to achieve high 3D imaging speed (10+ volumes per second) to avoid motion blur while also maintaining single cell resolution and reasonable field of view.Over the past decade, light sheet microscopy has emerged as a promising technique with great spatial resolution and reduced phototoxicity. Swept, confocally-aligned, planar excitation (SCAPE) microscopy, a single objective light sheet modality developed by Hillman lab, has the advantage of an open top geometry and fast 3D imaging speed. In this proposal, I detail my work towards imaging and tracking the whole C. elegans nervous system at cellular resolution using SCAPE and the NeuroPAL strain. The first chapter introduces fundamental concepts that link the microscopy field with the C. elegans community. The second chapter involves building a new SCAPE system that incorporates new optical components and a high-speed intensified camera. The goal is to construct a workhorse system capable of capturing real-time volumetric recordings with improved resolution. The improvements stem from an improved optical design as well as careful selection of magnification and scan parameters While the new imaging system is capable of capturing high-speed volumetric images of freely moving NeuroPAL worms with single-cell resolution, there is no suitable neuron tracking algorithm to robustly extract neural activities from the data. Indeed, the density of the neurons as well as the vigorous movement of the worm is unprecedented. Chapter 3 and 4 constitute two parts of a broader neuron tracking algorithm. In Chapter 3, I introduce an iterative neural network based algorithm for unsupervised 3D image registration. In Chapter 4, a Gaussian Mixture Model based algorithm is proposed that simulates the raw data as the mixture of 3D Gaussian functions. Chapter 5 is the finale where I integrate of all proposed imaging and tracking methods in recording neural activity from the whole nervous system in freely-behaving NeuroPAL worms. Three applications are demonstrated, which spans from whole nervous system recording to investigation of class-dependent ventral nerve cord motor neurons during locomotion. In Chapter 6, I report progress towards building the next-generation SCAPE with higher resolution/collection efficiency. A custom-designed zero working distance objective is demonstrated, which uses off-the-shelf objective with novel refractive-index-matched material to achieve high collection numerical aperture without sacrificing field of view (FOV).

Youth mental health impairments are a leading and growing cause of disability. Mental health deficits during childhood and adolescence often portend more serious illness later in life, and early intervention at time of symptom onset may be critical to ameliorating disease trajectories. However, despite their importance, the neural and developmental underpinnings of many psychiatric disorders are not well understood. This dissertation aims to improve our understanding of pediatric psychiatric disorders by harnessing the power of machine learning, large sample sizes, and distinct training and replication subsamples to robustly examine functional magnetic resonance imaging data in two large samples of youth. In Chapter 2, we review prior uses of machine learning in the psychiatric neuroimaging literature. We also develop a framework for evaluating machine learning applications in psychiatric neuroimaging, which we apply throughout this dissertation. In Chapter 3 (Study 1), we use several supervised and unsupervised machine learning techniques to probe functional neural correlates of obsessive-compulsive symptoms in a large, multi-site community sample of youth. We find that patterns of individual obsessive-compulsive symptoms are fairly stable across subsamples. Granular resting state functional connectivity patterns associated with those symptom dimensions are not reliable, but broader large-scale network patterns appear to be more stable across subsamples. In Chapter 4 (Study 2), we use a different large sample of youth to assess clinical, cognitive, and demographic factors associated with head motion during fMRI. Head motion is a known source of artifact in fMRI data, especially data collected from youth. Our findings suggest that head motion may be systematically associated with neuropsychiatric symptom severity, thus potentially confounding neuroimaging studies involving patient populations. Across studies, this dissertation highlights the need for reproducibility and replicability, with a focus on research transparency, code sharing, and pre-registration of analyses. We hope herein to provide a solid methodological foundation from which to build our understanding of the neural basis of pediatric psychiatric symptoms.

In 2023, breakthroughs achieved by large language models like ChatGPT have been transformative, revealing the hidden structures within natural language. This has enabled these models to reason and perform tasks with intelligence previously unattainable, a feat made possible by analyzing vast datasets. However, the domain of medical imaging—characterized by the high costs and intensive labor of data acquisition, along with the scarcity of data from pathological sources—presents unique challenges. Neuroimaging data, for instance, is marked by its high dimensionality, limited sample sizes, complex hierarchical and temporal structures, significant noise, and contextual variability. These obstacles are especially prevalent in methodologies like functional Magnetic Resonance Imaging (fMRI) and computer vision applications, where datasets are naturally sparse. Developing sophisticated methods to overcome these challenges is essential for maximizing the utility of imaging technologies and enhancing our understanding of neurological functions. Such advancements are critical for the creation of innovative diagnostic tools and therapeutic approaches for neurological and psychiatric conditions. The data from current set of non-invasive neuroimaging modalities is most often analyzed using classical statistical and machine learning methods. In this work we show that widely used machine learning methods for neural imaging data can be unified under a Bayesian perspective. We use this unifying view of probabilistic modeling techniques to further develop models and statistical inference methods to address the aforementioned challenges by leveraging substantial research developments in artificial intelligence i.e. deep learning, and probabilistic modeling methods over the last decade. In this work, we broaden the family of probabilistic models to encompass various prior structures,including discrete, hierarchical, and temporal elements. We derive efficient inference models using principled Bayesian inference and modern stochastic optimization and empirically demonstrate how the representational capacity of neural networks can be combined with principled probabilistic generative models to achieve state-of-the-art results on neuroimaging and computer vision datasets. The methods we develop are applicable to a diverse range of datasets beyond neuroimaging; for instance, we apply these probabilistic inference principles to improve movie and song recommendations, enhance object detection in computer vision models, and perform neural architecture search.

Our decisions reflect who we are and shape who we become. The field of decision-making and learning has revealed how past outcomes guide our future choices by updating value representations. However, we often make choices among options we never experienced before, and therefore their value is not immediately accessible. How do we navigate these choices? This dissertation posits that such decisions rely on memory mechanisms that facilitate generalization and inferential reasoning. Through a combination of behavioral experiments, computational modeling and neuroimaging measures, I explore the diverse ways in which memory mechanisms influence value-based decisions between options for which value is unknown. In Chapter One, I investigate how individuals assign value to unchosen alternatives and propose that memory creates an associative link between choice options, with consequences for later updating of value. Through a series of five experiments, I demonstrate an inverse relationship between the valuation of unchosen options and the direct learning about chosen options, and show that this inverse inference of value is related to memory of the decision itself. In Chapter Two, I manipulate the associative link between choice options using a well-established memory manipulation technique and observe a reduction in the inverse inference of unchosen options. This provides further evidence for the causal role of associative memory in the inferential effect. Additionally, I introduce a novel policy-gradient model incorporating memory components that offers the best explanation for observed behaviors. Lastly, in Chapter Three, I present behavioral and neuroimaging evidence supporting the influence of conceptual knowledge in value-based decisions involving entirely new choice options. I show that people use existing category knowledge to group similar items together, enabling value extraction at a category level and generalization to novel items. Overall, this dissertation underscores the fundamental role of memory in shaping the construction and use of value to guide choice. It emphasizes the adaptive and flexible nature of memory, showing how it combines past experiences to guide future actions.

Neurological disorders present a significant challenge in global health. With the increasing availability of imaging datasets and the development of precise machine learning models, early and accurate diagnosis of neurological conditions is a promising and active area of research. However, several characteristic factors in neurology domains, such as heterogeneous imaging, inaccurate labels, or limited data, act as bottlenecks in using deep learning on clinical neuroimaging. Given these circumstances, this dissertation attempts to provide a guideline, proposing several methods and showcasing successful implementations in broad neurological conditions, including epilepsy and neurodegeneration. Methodologically, a particular focus is on comparing a two-dimensional approach as opposed to three-dimensional neural networks. In most clinical domains of neurological disorders, data are scarce and signals are weak, discouraging the use of 3D representation of raw scan data. This dissertation first demonstrates competitive performances with 2D models in tuber segmentation and AD comorbidity detection. Second, the potentials of ensemble learning are explored, further justifying the use of 2D models in the identification of neurodegeneration. Lastly, CleanNeuro is introduced in the context of 2D classification, a novel algorithm for denoising the datasets prior to training. CleanNeuro, on top of 2D classification and ensemble learning, demonstrates the feasibility of accurately classifying patients with comorbid AD and cerebral amyloid angiopathy from AD controls. Methods presented in this dissertation may serve as exemplars in the study of neurological disorders using deep learning and clinical neuroimaging. Clinically, this dissertation contributes to improving automated diagnosis and identification of regional vulnerabilities of several neurological disorders on clinical neuroimaging using deep learning. First, the classification of patients with Alzheimer’s disease from cognitively normal group demonstrates the potentials of using positron emission tomography with tau tracers as a competitive biomarker for precision medicine. Second, the segmentation of tubers in patients with tuberous sclerosis complex proves a successful 2D modeling approach in quantifying neurological burden of a rare yet deadly disease. Third, the detection of comorbid pathologies from patients with Alzheimer’s disease is analyzed and discussed in depth. Based on prior findings that comorbidities of Alzheimer’s disease affect the brain structure in a distinctive pattern, this dissertation proves for the first time the effectiveness of using deep learning on the accurate identification of comorbid pathology in vivo. Leveraging postmortem neuropathology as ground truth labels on top of the proposed methods records competitive performances in comorbidity prediction. Notably, this dissertation discovers that structural magnetic resonance imaging is a reliable biomarker in differentiating the comorbid cereberal amyloid angiopathy from Alzheimer’s disease patients. The dissertation discusses experimental findings on a wide range of neurological disorders, including tuberous sclerosis complex, dementia, and epilepsy. These results contribute to better decision-making on building neural network models for understanding and managing neurological diseases. With the thorough exploration, the dissertation may provide valuable insights that can push forward research in clinical neurology.