Development and characterization of deep learning techniques for neuroimaging data

Deep learning methods are extremely promising machine learning tools to analyze neuroimaging data. However, their potential use in clinical settings is limited because of the existing challenges of applying these methods to neuroimaging data. In this study, first a data leakage type caused by slice-level data split that is introduced during training and validation of a 2D CNN is surveyed and a quantitative assessment of the model’s performance overestimation is presented. Second, an interpretable, leakage-fee deep learning software written in a python language with a wide range of options has been developed to conduct both classification and regression analysis. The software was applied to the study of mild cognitive impairment (MCI) in patients with small vessel disease (SVD) using multi-parametric MRI data where the cognitive performance of 58 patients measured by five neuropsychological tests is predicted using a multi-input CNN model taking brain image and demographic data. Each of the cognitive test scores was predicted using different MRI-derived features. As MCI due to SVD has been hypothesized to be the effect of white matter damage, DTI-derived features MD and FA produced the best prediction outcome of the TMT-A score which is consistent with the existing literature. In a second study, an interpretable deep learning system aimed at 1) classifying Alzheimer disease and healthy subjects 2) examining the neural correlates of the disease that causes a cognitive decline in AD patients using CNN visualization tools and 3) highlighting the potential of interpretability techniques to capture a biased deep learning model is developed. Structural magnetic resonance imaging (MRI) data of 200 subjects was used by the proposed CNN model which was trained using a transfer learning-based approach producing a balanced accuracy of 71.6%. Brain regions in the frontal and parietal lobe showing the cerebral cortex atrophy were highlighted by the visualization tools.

Qualification of optical fiber sensors for the structural health monitoring of aerospace structures

In recent years, composite materials have revolutionized the design of many structures. Their superior mechanical properties and light weight make composites convenient over traditional metal structures for many applications. However, composite materials are susceptible to complex and challenging to predict damage behaviors due to their anisotropy nature. Therefore, structural Health Monitoring (SHM) can be a valuable tool to assess the damage and understand the physics underneath. Distributed Optical Fiber Sensors (DOFS) can be used to monitor several types of damage in composites. However, their implementation outside academia is still unsatisfactory. One of the hindrances is the lack of a rigorous methodology for uncertainty quantification, which is essential for the performance assessment of the monitoring system. The concept of Probability of Detection (POD) must function as the guiding light in this process. However, precautions must be taken since this tool was established for Non-Destructive Evaluation (NDE) rather than Structural Health Monitoring (SHM). In addition, although DOFS have been the object of numerous studies, a well-established POD methodology for their performance assessment is still missing. This thesis aims to develop a methodology to produce POD curves for DOFS in composite materials. The problem is analyzed considering several critical points, such as the strain transfer characterizing the DOFS and the development of an experimental and model-assisted methodology to understand the parameters that affect the DOFS performance.

Structural and stratigraphic control on hypogene dissolution in carbonate rocks: case studies of cave analogues and implications for karst reservoirs

The aim of this research is to improve the understanding of the factors that control the formation of karst porosity in hypogene settings and its associated patterns of void-conduit networks. Subsurface voids created by hypogene dissolution may span from few microns to decametric tubes providing interconnected conduit systems and forming highly anisotropic permeability domains in many reservoirs. Characterizing the spatial-morphological organization of hypogene karst is a challenging task that has dramatic implications for the applied industry, given that only partial data can be acquired from the subsurface by indirect techniques. Therefore, two outcropping cave analogues are examined: the Cavallone-Bove Cave in the Majella Massif (Italy), and the karst systems of the Salitre Formation (Brazil). In the latter, a peculiar example of hypogene speleogenesis associated with silicification has been studied, providing an analogue of many karstified reservoirs hosted in cherts or cherty-carbonates within mixed sedimentary sequences. The first part of the thesis is focused on the relationships between fracture patterns and flow pathways in deformed units in: 1) a fold-and-thrust setting (Majella Massif); 2) a cratonic block (Brazil). These settings represent potential playgrounds for the migration and accumulation of geofluids, where hypogene conduits may affect flow pathways, fluid storage, and reservoir properties. The results indicate that localized deformation producing cross-formational fracture zones associated with anticline hinges or fault damage zones is critical for hypogene fluid migration and karstification. The second part of the thesis deals with the multidisciplinary study of hydrothermal silicification and hypogene dissolution in Calixto Cave (Brazil). Petrophysical analyses and a geochemical characterization of silica deposits are used to unravel the spatial-morphological organization of the conduit system and its speleogenesis. The novel results obtained from this cave shed new light on the relationship between hydrothermal silicification, hypogene dissolution and the development of multistorey cave systems in layered carbonate-siliciclastic sequences.

Cardiac functional and structural abnormalities in a mouse model of CDKL5 deficiency disorder

CDKL5 (cyclin-dependent kinase-like 5) deficiency disorder (CDD) is a rare and severe neurodevelopmental disease that mostly affects girls who are heterozygous for mutations in the X-linked CDKL5 gene. The lack of CDKL5 protein expression or function leads to the appearance of numerous clinical features, including early-onset seizures, marked hypotonia, autistic features, and severe neurodevelopmental impairment. Mouse models of CDD, Cdkl5 KO mice, exhibit several behavioral phenotypes that mimic CDD features, such as impaired learning and memory, social interaction, and motor coordination. CDD symptomatology, along with the high CDKL5 expression levels in the brain, underscores the critical role that CDKL5 plays in proper brain development and function. Nevertheless, the improvement of the clinical overview of CDD in the past few years has defined a more detailed phenotypic spectrum; this includes very common alterations in peripheral organ and tissue function, such as gastrointestinal problems, irregular breathing, hypotonia, and scoliosis, suggesting that CDKL5 deficiency compromises not only CNS function but also that of other organs/tissues. Here we report, for the first time, that a mouse model of CDD, the heterozygous Cdkl5 KO (Cdkl5 +/-) female mouse, exhibits cardiac functional and structural abnormalities. The mice also showed QTc prolongation and increased heart rate. These changes correlate with a marked decrease in parasympathetic activity to the heart and in the expression of the Scn5a and Hcn4 voltage-gated channels. Moreover, the Cdkl5 +/- heart shows typical signs of heart aging, including increased fibrosis, mitochondrial dysfunctions, and increased ROS production. Overall, our study not only contributes to the understanding of the role of CDKL5 in heart structure/function but also documents a novel preclinical phenotype for future therapeutic investigation.

Study of charge, spin, and structural orderings in quantum materials using nuclei and muons as local probes

Quantum Materials are many body systems displaying emergent phenomena caused by quantum collective behaviour, such as superconductivity, charge density wave, fractional hall effect, and exotic magnetism. Among quantum materials, two families have recently attracted attention: kagome metals and Kitaev materials. Kagome metals have a unique crystal structure made up of triangular lattice layers that are used to form the kagome layer. Due to superconductivity, magnetism, and charge ordering states such as the Charge Density Wave (CDW), unexpected physical phenomena such as the massive Anomalous Hall Effect (AHE) and possible Majorana fermions develop in these materials. Kitaev materials are a type of quantum material with a unique spin model named after Alexei Kitaev. They include fractional fluctuations of Majorana fermions and non-topological abelian anyons, both of which might be used in quantum computing. Furthermore, they provide a realistic framework for the development of quantum spin liquid (QSL), in which quantum fluctuations produce long-range entanglements between electronic states despite the lack of classical magnetic ordering. In my research, I performed several nuclear magnetic resonance (NMR), nuclear quadrupole resonance (NQR), and muon spin spectroscopy (µSR) experiments to explain and unravel novel phases of matter within these unusual families of materials. NMR has been found to be an excellent tool for studying these materials’ local electronic structures and magnetic properties. I could use NMR to determine, for the first time, the structure of a novel kagome superconductor, RbV3Sb5, below the CDW transition, and to highlight the role of chemical doping in the CDW phase of AV3Sb5 superconductors. µSR has been used to investigate the effect of doping on kagome material samples in order to study the presence and behaviour of an anomalous phase developing at low temperatures and possibly related to time-reversal symmetry breaking.

Advanced X-ray techniques for the structural characterisation of Prussian Blue Analogue cathode materials

The thesis is dedicated to the implementation of advanced x-ray-based techniques for the investigation of the battery systems, more predominantly, the cathode materials. The implemented characterisation methods include synchrotron based x-ray absorption spectroscopy, powder x-ray diffraction, 2-dimensional x-ray fluorescence, full field transmission soft x-ray microscopy, and laboratory x-ray photoelectron spectroscopy. The research highlights the different areas of expertise for each described method, in terms of material characterisation, exploring their complementarities and intersections. The results are focused over manganese hexacyanoferrate and partially Ni substituted manganese hexacyanoferrate, through both organic and aqueous battery systems. In aqueous system, the modification of cathode composition has been observed with various techniques, indicating to the processes occurring in bulk, surface, locally or in long-range, including with the speciation by 2-dimensional scanning, and the time-resolution, by the implementation of the operando measurements. In organic media, the inhomogenisation of the cathode material during the aging process was investigated by the development of the special image treatment procedure for the maps, obtained from the transmission soft x-ray microscopy. It worth mentioning, that apart from the combination of the outcomes from the various x-ray measurements, the exploration of the new capabilities was also conducted, namely, probing the oxidation state of the element with the synchrotron-based 2-dimensional x-ray fluorescence technique, which, generally, with conventional set up, is not possible to achieve. The results and methodology from this thesis can, of course, be generalised on the characterisation of the other battery systems, and not only, as the x-ray techniques are one of the most informative and sophisticated methods for advanced structural investigation of the materials.