Interpretable Convolutional Neural Networks for Decoding and Analyzing Neural Time Series Data

Machine learning is widely adopted to decode multi-variate neural time series, including electroencephalographic (EEG) and single-cell recordings. Recent solutions based on deep learning (DL) outperformed traditional decoders by automatically extracting relevant discriminative features from raw or minimally pre-processed signals. Convolutional Neural Networks (CNNs) have been successfully applied to EEG and are the most common DL-based EEG decoders in the state-of-the-art (SOA). However, the current research is affected by some limitations. SOA CNNs for EEG decoding usually exploit deep and heavy structures with the risk of overfitting small datasets, and architectures are often defined empirically. Furthermore, CNNs are mainly validated by designing within-subject decoders. Crucially, the automatically learned features mainly remain unexplored; conversely, interpreting these features may be of great value to use decoders also as analysis tools, highlighting neural signatures underlying the different decoded brain or behavioral states in a data-driven way. Lastly, SOA DL-based algorithms used to decode single-cell recordings rely on more complex, slower to train and less interpretable networks than CNNs, and the use of CNNs with these signals has not been investigated. This PhD research addresses the previous limitations, with reference to P300 and motor decoding from EEG, and motor decoding from single-neuron activity. CNNs were designed light, compact, and interpretable. Moreover, multiple training strategies were adopted, including transfer learning, which could reduce training times promoting the application of CNNs in practice. Furthermore, CNN-based EEG analyses were proposed to study neural features in the spatial, temporal and frequency domains, and proved to better highlight and enhance relevant neural features related to P300 and motor states than canonical EEG analyses. Remarkably, these analyses could be used, in perspective, to design novel EEG biomarkers for neurological or neurodevelopmental disorders. Lastly, CNNs were developed to decode single-neuron activity, providing a better compromise between performance and model complexity.

Relative influence of gaze and static arm position in the Superior Parietal Lobule of the macaque monkey during reaching

The superior parietal lobule (SPL) of macaques is classically described as an associative cortex implicated in visuospatial perception, planning and control of reaching and grasping movements (De Vitis et al., 2019; Galletti et al., 2003, 2018, 2022; Fattori et al., 2017; Hadjidimitrakis et al., 2015). These processes are the result of the integration of signals related to different sensory modalities. During a goal-directed action, eye and limb information are combined to ensure that the hand is transported at the gazed target location and the arm is maintained steady in the final position. The SPL areas V6A, PEc and PE contain cells sensitive to the direction of gaze and limb position but less is known about the degree of independent encoding of these signals. In this thesis, we evaluated the influence of eye and arm position information upon single neuron activity of areas V6A, PEc and PE during the holding period after the execution of arm reaching movement, when the gaze and hand are both still at the reach target. Two male macaques (Macaca fascicularis) performed a reaching task while single unit activity was recorded from areas V6A, PEc and PE. We found that neurons in all these areas were modulated by eye and static arm positions with a joint encoding of gaze and somatosensory signals in V6A and PEc and a mostly separate processing of the two signals in PE. The elaboration of this information reflects the functional gradient found in the SPL with the caudal sector characterized by visuo-somatic properties in comparison to the rostral sector dominated by somatosensory signals. This evidence well agree also with the recent reallocation of areas V6A and PEc in Brodmann’s area 7 depending on their similar structural and functional features with respect to PE belonging to Brodmann’s area 5 (Gamberini et al., 2020).

Pathogenetic mechanisms of gastrointestinal symptoms in a mouse model of Fabry disease: insights on the microbiota-gut-brain axis

Fabry disease (FD), X-linked metabolic disorder caused by a deficiency in α-galactosidase A activity, leads to the accumulation of glycosphingolipids, mainly Gb3 and lyso-Gb3, in several organs. Gastrointestinal (GI) symptoms are among the earliest and most common, strongly impacting patients’ quality of life. However, the origin of these symptoms and the exact mechanisms of pathogenesis are still poorly understood, thus the pressing need to improve their knowledge. Here we aimed to evaluate whether a FD murine model (α-galactosidase A Knock-Out) captures the functional GI issues experienced by patients. In particular, the potential mechanisms involved in the development and maintenance of GI symptoms were explored by looking at the microbiota-gut-brain axis involvement. Moreover, we sought to examine the effects of lyso-Gb3 on colonic contractility and the intestinal epithelium and the enteric nervous system, which together play important roles in regulating intestinal ion transport and fluid and electrolyte homeostasis. Fabry mice revealed visceral hypersensitivity and a diarrhea-like phenotype accompanied by anxious-like behavior and reduced locomotor activity. They reported also an imbalance of SCFAs and an early compositional and functional dysbiosis of the gut microbiota, which partly persisted with advancing age. Moreover, overexpression of TRPV1 was found in affected mice, and partial alteration of TRPV4 and TRPA1 as well, identifying them as possible therapeutic targets. The Ussing chamber results after treatment with lyso-Gb3 showed an increase in Isc (likely mediated by HCO3- ions movement) which affects neuron-mediated secretion, especially capsaicin- and partly veratridine-mediated. This first characterization of gut-brain axis dysfunction in FD mouse provides functional validation of the model, suggesting new targets and possible therapeutic approaches. Furthermore, lyso-Gb3 is confirmed to be not only a marker for the diagnosis and follow-up of FD but also a possible player in the alteration of the FD colonic ion transport process.

Role of neuroinflammation in the pathophysiology of CDKL5 deficiency disorder

Cyclin-dependent kinase-like 5 (CDKL5) deficiency disorder (CDD), a rare neurodevelopmental disease caused by mutations in the X-linked CDKL5 gene, is characterized by early-onset epilepsy, intellectual disability, and autistic features. To date, little is known about the etiology of CDD and no therapies are available. When overactivated in response to neuronal damage and genetic or environmental factors, microglia – the brain macrophages – cause damage to neighboring neurons by producing neurotoxic factors and pro-inflammatory molecules. Importantly, overactivated microglia have been described in several neurodegenerative and neurodevelopmental disorders, suggesting that active neuroinflammation may account for the compromised neuronal survival and/or brain development observed in these pathologies. Recent evidence shows a subclinical chronic inflammatory status in plasma from CDD patients. However, it is unknown whether a similar inflammatory status is present in the brain of CDD patients and, if so, whether it plays a causative or exacerbating role in the pathophysiology of CDD. Here, we show evidence of a chronic microglia overactivation status in the brain of Cdkl5 KO mice, characterized by alterations in microglial cell number/morphology and increased pro-inflammatory gene expression. We found that the neuroinflammatory process is already present in the postnatal period in Cdkl5 KO mice and worsens during aging. Remarkably, by restoring microglia alterations, treatment with luteolin, a natural anti-inflammatory flavonoid, promotes neuronal survival in the brain of Cdkl5 KO mice since it counteracts hippocampal neuron cell death and protects neurons from NMDA-induced excitotoxic damage. In addition, through the restoration of microglia alterations, luteolin treatment also increases hippocampal neurogenesis and restores dendritic spine maturation and dendritic arborization of hippocampal and cortical pyramidal neurons in Cdkl5 KO mice, leading to improved behavioral performance. These findings highlight new insights into the CDD pathophysiology and provide the first evidence that therapeutic approaches aimed at counteracting neuroinflammation could be beneficial in CDD.

Quantitative susceptibility mapping as biomarker of neurodegeneration: methodological models and clinical applications

Quantitative Susceptibility Mapping (QSM) is an advanced magnetic resonance technique that can quantify in vivo biomarkers of pathology, such as alteration in iron and myelin concentration. It allows for the comparison of magnetic susceptibility properties within and between different subject groups. In this thesis, QSM acquisition and processing pipeline are discussed, together with clinical and methodological applications of QSM to neurodegeneration. In designing the studies, significant emphasis was placed on results reproducibility and interpretability. The first project focuses on the investigation of cortical regions in amyotrophic lateral sclerosis. By examining various histogram susceptibility properties, a pattern of increased iron content was revealed in patients with amyotrophic lateral sclerosis compared to controls and other neurodegenerative disorders. Moreover, there was a correlation between susceptibility and upper motor neuron impairment, particularly in patients experiencing rapid disease progression. Similarly, in the second application, QSM was used to examine cortical and sub-cortical areas in individuals with myotonic dystrophy type 1. The thalamus and brainstem were identified as structures of interest, with relevant correlations with clinical and laboratory data such as neurological evaluation and sleep records. In the third project, a robust pipeline for assessing radiomic susceptibility-based features reliability was implemented within a cohort of patients with multiple sclerosis and healthy controls. Lastly, a deep learning super-resolution model was applied to QSM images of healthy controls. The employed model demonstrated excellent generalization abilities and outperformed traditional up-sampling methods, without requiring a customized re-training. Across the three disorders investigated, it was evident that QSM is capable of distinguishing between patient groups and healthy controls while establishing correlations between imaging measurements and clinical data. These studies lay the foundation for future research, with the ultimate goal of achieving earlier and less invasive diagnoses of neurodegenerative disorders within the context of personalized medicine.

Directly training spiking neural networks for cyber-physical systems: from supervised to reinforcement learning

Spiking Neural Networks (SNNs) are bio-inspired Artificial Neural Networks (ANNs) utilizing discrete spiking signals, akin to neuron communication in the brain, making them ideal for real-time and energy-efficient Cyber-Physical Systems (CPSs). This thesis explores their potential in Structural Health Monitoring (SHM), leveraging low-cost MEMS accelerometers for early damage detection in motorway bridges. The study focuses on Long Short-Term SNNs (LSNNs), although their complex learning processes pose challenges. Comparing LSNNs with other ANN models and training algorithms for SHM, findings indicate LSNNs' effectiveness in damage identification, comparable to ANNs trained using traditional methods. Additionally, an optimized embedded LSNN implementation demonstrates a 54% reduction in execution time, but with longer pre-processing due to spike-based encoding. Furthermore, SNNs are applied in UAV obstacle avoidance, trained directly using a Reinforcement Learning (RL) algorithm with event-based input from a Dynamic Vision Sensor (DVS). Performance evaluation against Convolutional Neural Networks (CNNs) highlights SNNs' superior energy efficiency, showing a 6x decrease in energy consumption. The study also investigates embedded SNN implementations' latency and throughput in real-world deployments, emphasizing their potential for energy-efficient monitoring systems. This research contributes to advancing SHM and UAV obstacle avoidance through SNNs' efficient information processing and decision-making capabilities within CPS domains.