e-Repositori UPF :: Grau en Enginyeria Biomèdica. Treballs de fi de grau

Recent Submissions

Optimizing transcranial pulse stimulation for neuropathic pain: a computational planning approach
(2025) Urgel Cantalejo, Keysha
Transcranial Pulsed Stimulation (TPS) is a non-invasive neuromodulation technique that uses focused ultrashort ultrasound pulses to stimulate specific regions of the brain. By delivering controlled acoustic pulses through the skull, TPS enables targeted stimulation with high spatial precision, while minimizing discomfort and side effects. This thesis explores the optimization of TPS therapy for treating neuropathic pain, a chronic and debilitating condition caused by abnormal nervous system activity. Traditional treatments, such as medication or invasive interventions, often yield limited results or pose significant risks. In contrast, TPS offers a non-invasive alternative by targeting, in this case, the dorsal anterior cingulate cortex (dACC), a region involved in the emotional processing of pain. By modulating this area, TPS has the potential to alleviate the affective component of pain and serve as a predictor for the success of more invasive treatments, such as Deep Brain Stimulation (DBS). However, current TPS procedures rely primarily on visual positioning of the transducer, without assessing whether the ultrasound energy is optimally delivered to the target area. This thesis addresses that limitation by proposing a computational method to optimize the transducer’s placement using patient-specific CT data. Heuristic metrics are used to quickly evaluate potential stimulation paths, while high-resolution simulations validate the best candidates.
Optimizing the research ethics committee’s assessment of biomedical research proposals
(2025) Vázquez Masana, Ruth
Research Ethics Committees (REC) are under significant pressure as the volume of Biomedical Research Proposals (BRP) grows, the escalating complexity of scientific knowledge due to the advent of Artificial Intelligence, and the proliferation of immature proposals submitted for evaluation. This study introduces Pre-Val-AI-THICS, an artificial intelligence-based prevalidation tool and an ontology, designed to optimize the review of BRP by REC at Hospital Clínic de Barcelona (HCB). The work employed a sequential methodology comprising an initial analysis of REC evaluation workflows followed by the design and feasibility assessment of the tool during a hackathon event. It culminated in the development of a proof of concept integrating both automated extraction of contents and AI-assisted review capabilities. The Pre-Val-AI-THICS system consists of two primary components: a Python-based module utilizing regular expressions for BRP content extraction and verifying compliance with the HCB’s standardized template, and an AI module leveraging Chain-of-Thought prompt engineering with Llama 3.2 3B, to detect and suggest corrections related to data management and study design inconsistencies. Complementing these components, a preliminary BRP ontology was constructed using Protégé to formalize key concepts and relationships inherent in BRP evaluation. It was demonstrated that Pre-Val-AI-THICS effectively identified recurrent data management and structural errors, providing actionable recommendations to enhance protocol quality prior to REC submission. While the ontology remains in a nascent stage, it is envisioned as a foundation for the future development of a knowledge graph supporting standardized representation and evaluation of BRP across diverse committees. Overall, the findings underscore the potential of AI-driven prevalidation tools to streamline REC workflows, reduce administrative burden, and improve the consistency and rigor of ethical assessments. Further research will focus on advancing ontology-driven navigation and interoperability to enable comprehensive, semantically enriched access to BRP content and evaluation criteria.
Development of a multimodal strategy to investigate speech processing mechanisms in infants
(2025) Valls Santafé, Silvia
Understanding how infants process speech signal is a question of interest in neuroscience, because it provides insight into the early language acquisition and cognition development. However, investigating these mechanisms requires neuroimaging tools capable of capturing brain activity, adapted to the specific challenges of measuring this population. This Bachelor’s thesis contributes to a broader project that aims to uncover speech processing mechanisms in 4-5-month-old infants, presenting the development of a multimodal strategy that integrates electroencephalography (EEG), functional Diffusion Correlation Spectroscopy (fDCS) and functional Near-Infrared Spectroscopy (fNIRS) to acquire data by characterizing the neuronal activity and its relation to oxygen metabolism (i.e., energetic demand and oxygen consumption). To do so, this project aims to develop a multimodal system that combines optical techniques (fNIRS and fDCS) with EEG. Optical techniques would provide hemodynamic response, while EEG will quantify the electrical activity of the neurons enabling acquisition with both high temporal and spatial resolution. To implement this approach, a custom pipeline combining Python with 3D modelling was developed to automate the generation of sensor configuration (EEG electrodes, fNIRS and fDCS optodes) while respecting anatomical and technical constraints. Moreover, it also addresses the challenge of reducing the attrition rate caused by poor signal quality due to hair related issues in fDCS measurements. To address this challenge, a novel mechanism was developed based on pressure adjustment. Following multiple iterations, some models were prototyped through 3D printing and CNC (Computer Numerical Control) turning to ensure mechanical robustness and usability. The final system has the potential to improve contact across diverse hair types, to reduce setup time and to minimize participant exclusion. While further improvements are needed before deployment, the system represents a meaningful step toward enabling simultaneous acquisition of EEG, fNIRS and fDCS data in early infancy.
Developing a longitudinal framework to explore how deep brain stimulation parameters influence cognitive decline in Parkinson’s disease
(2025) Dalmau Baltà, Marta
Deep Brain Stimulation (DBS) of the subthalamic nucleus (STN) is an established treatment for motor symptoms in Parkinson’s disease (PD), yet its e!ects on cognitive function remain less understood. This study develops a workflow within Lead-DBS environment to automatically compute volumes of tissue activated (VTAs) across multiple postoperative timepoints. In addition, it analyzes these data by assessing whether stimulation parameters are associated with cognitive decline following STN-DBS. A structured dataset of 64 PD patients who underwent STN-DBS was used, which included a complete cognitive performance assessed with the Mattis Dementia Rating Scale (MDRS) and stimulation data across five postoperative time points. The developed pipeline enabled automatic VTA generation from stimulation parameters and organized patient data in a structure compatible with Lead-DBS. This allowed for the computation of stimulation overlap with the specific STN subregions at each timepoint. To explore potential associations with cognitive change, patients were clustered based on their longitudinal Mattis trajectories. Two distinct subgroups were identified: one showing stable or improved cognitive scores, and another exhibiting progressive decline. Linear mixed-e!ects models confirmed a significant divergence in cognitive outcomes between these groups over time. Although stimulation overlap with STN subregions did not di!er significantly between clusters, the declining group showed a consistent trend toward greater overlap in the motor and associative territories. Exploratory sweetspot analyses gave further information regarding voxel-wise distribution of the sweet and sour spots. Despite limitations such as a relatively small sample size and the absence of genetic data, this study demonstrates the feasibility of longitudinal VTA analysis using Lead DBS. Although no robust evidence between stimulation and cognitive decline was found in this cohort, subtle trends suggest a potential relationship that cannot be completely discarded. Future work should integrate this approach with more ad- vanced voxel-wise and connectivity-based methods to better understand the neural mechanisms underlying cognitive outcomes in DBS-treated patients.
Deep learning segmentation for morphological assessment of optic nerve integrity in optic neuritis
(2025) Sánchez Ulloa, Helena
Studying the optic nerve in Multiple Sclerosis (MS) patients plays a crucial role in early diagnosis and non-invasive monitoring of disease progression, as Optic Neuritis (ON) is a frequent and often early manifestation of MS. Magnetic Resonance Imaging (MRI) is the technique of choice to assess the integrity of the optic nerve. By examining the presence of lesions in the optic nerve, clinicians can obtain insights into the progression of the disease and its impact on the central nervous system over time. The goal of this bachelor’s thesis is to generate, from MRI, detailed optic nerve profiles and morphological assessments, allowing for a comparison across different patients. These profiles can help to identify the presence of lesions, track disease progression and predict clinical outcomes. To achieve this, a cohort of subjects with MS, with and without ON, and healthy subjects will be analyzed. Deep learning models will be trained using 3D T1-weighted MRI scans from this dataset to segment the optic nerve. The assessment of optic nerve integrity will be performed by calculating the T1/T2 ratio, enabling precise detection and analysis of ON lesions and allowing for a comparison of these profiles with those of the control patients.
Rebalancing the depressed brain: a wholebrain computational study on the effects of external perturbations in psilocybin and escitalopram treatments
(2024) Socoró Garrigosa, Marcel
Serotonergic psychedelics like psilocybin have been proposed as a promising avenue in the treatment of major depressive disorder (MDD). Contrary to selective serotonin reuptake inhibitors (SSRI) like escitalopram, psychedelics work through 5-HT2AR agonism, which opens a window of plasticity for psychotherapy- or neurostimulation-combined therapies. Here, we used a trial comparing psilocybin and escitalopram treatments for MDD to assess in silico the effect of external perturbations after pharmacological intervention. The trial included resting state fMRI scans before and one day after treatment. We built whole-brain models by fitting the data to capture the underlying causal brain mechanisms in generative effective connectivity. Then, we applied a perturbation protocol following the Dynamic Sensitivity Analysis paradigm, simulating the effects of external stimulations in parcellated regions. We evaluated the impact of perturbations by assessing regional susceptibility to change and effectivity to drive a healthy transition (both defined using static functional connectivity). We show that susceptibility is enhanced by psilocybin and reduced by escitalopram, evidencing an opening of a window of plasticity by psilocybin. However, healthy transitions are similarly achieved after both treatments, suggesting that escitalopram manages to improve perturbation effects despite differences in functional hierarchies compared to psilocybin. We elaborate that longer-term data and measures capturing the temporal richness of functional repertoires might be needed to complement these findings. Finally, we demonstrate the benefits of multi-site versus single-site perturbations and that the amygdala and the nucleus accumbens are the best perturbation targets to drive healthy transitions. Overall, the present work contributes to prior knowledge on how in silico perturbations can help in the treatment of MDD, serving as a foundation for larger computational studies.
Enhanced prioritization and reporting for coronary artery disease diagnosis
(2024) Ferrer Beltran, Eva
Coronary artery disease (CAD) remains a significant cause of mortality, particularly in industrialized nations, contributing to a significant portion of deaths worldwide. Timely diagnosis and treatment are critical for mitigating its high morbidity and mortality rates. However, current diagnostic protocols in hospitals often suffer from inefficiencies and lengthy processes. This Bachelor’s thesis aimed to optimize the CAD diagnosis workflow within the radiology department of Hospital de la Santa Creu i Sant Pau. The focus was on implementing methodologies for automatic case reporting and patient prioritization, specifically in the scheduling of coronary computed tomography angiography (CCTA) acquisitions and determining the order of CCTA reporting. This optimization was achieved in collaboration with clinical and technological experts from the institution. The implemented algorithms are rule-based, guided by clinical criteria from the literature and physicians. The prioritization of patients ensures that severe cases receive prompt attention, aiming to reduce diagnosis time. Additionally, the automatic generation of preliminary reports integrates clinical information and findings extracted from CCTA images through image analysis. Implementing the optimized workflow for CAD diagnosis through simulation experiments demonstrated a potential reduction in diagnosis time for more severe cases and increased clinician satisfaction with the CAD diagnosis workflow and workload. This study highlights the transformative potential of automatic approaches in streamlining CAD diagnosis, offering promise for improved patient care outcomes in clinical practice.
Multi-scale model for simulating thrombus formation and anticoagulant treatment
(2024) Ribera Pascual, Marina
Thrombosis is the localised clotting of blood and has a significant impact on medicine. Thrombi are formed by the activation of a series of events in the coagulation cascade. One way to prevent thrombus formation is by administering anticoagulants, which are drugs that target specific proteins in this cascade. In this work, a coagulation cascade model was developed. It provides a valuable tool for understanding the complex dynamics of thrombus formation and the role of the coagulation factors. Specifically, this model describes the temporal behaviour of several key components, including the tissue factor/factor VII complex, factor XI, factor IX, factor X, prothrombin, thrombin, factor VIII, factor V, protein C, peptide P, inactive and active fibrinogen, fibrin matrix and three intermediate complexes. The increase in viscosity that occurs as the thrombus develops is also described. The influence of tissue factor, factor XI and prothrombin on thrombus formation was also analysed through a sensitivity analysis. Prothrombin was found to have a major influence, as it is the precursor of thrombin, a crucial factor in the coagulation cascade. The coagulation cascade model was coupled to a fluid simulation of an idealised aneurysm, providing a comprehensive view of the spatial-temporal distribution of the coagulation factors based on the integration of residence time. The mapping of coagulation factors allows for a more realistic representation of physiological conditions. Inside the aneurysm, higher concentrations of coagulation factors were observed, as well as the presence of vortices, suggesting thrombus formation inside the aneurysm. In addition, the effect of the anticoagulant warfarin was incorporated into the coagulation cascade model. While warfarin successfully lowered the concentrations of the coagulation factors and delayed their activation, it was insufficient to prevent thrombus formation, as the same concentration was achieved as without anticoagulant.
A novel biophysical whole-brain model explains power spectrum alterations of serotonergic psychedelics
(2024) Cases Gendra, Jan
Background – Psychedelics hold great potential to treat various mental disorders, yet their neurobiological mechanisms remain unclear. Recent mechanistic models have provided valuable insights into the impact of serotonergic psychedelics on brain dynamics. However, these models focus mainly on macroscale brain activity and provide limited information on the psychedelics’ mechanisms at the neural population level. Methods – We provide a novel mechanistic explanation of the well-studied power spectrum alterations in spontaneous cortical activity observed under serotonergic psychedelics. We combine a physiologically grounded whole-brain model optimised with multimodal neuroimaging of healthy human participants with neurotransmitter data from positron emission tomography (PET) of the serotonin 2A receptor (5-HT2AR) density map. Building upon the recent laminar neural mass modelling (LaNMM) framework (Sanchez-Todo et al., 2023), the whole-brain model simulates multiband activity and electrophysiological measurements of the cortical columns, where the 5-HT2AR density controls the average synaptic gain of excitatory connections to layer 5 pyramidal neural populations, known to be rich in 5-HT2ARs. Results – Our findings suggest that the decrease in spontaneous cortical oscillatory power in the alpha band and increase in the gamma band are mainly influenced by the 5-HT2AR-mediated excitation of deep-layer pyramidal cells. These findings explain the functional effects of 5-HT2AR activation with psychedelics and allow us to propose a novel whole-brain biologically-informed explanation for this phenomenon detailed at the mesoscale and based on NMMs. Perspectives – This model provides valuable insight into the mechanistic underpinnings of psychedelic action in the brain and could be employed to investigate the neuromodulatory potential of psychedelics in re-establishing healthy brain dynamics in mental disorders.
Prediction of seizure onset zone in epilepsy patients via a network coupling measure
(2024) Elizondo Urrutia, Saioa
Epilepsy, a chronic neurological disorder characterized by recurrent seizures, affects millions globally. For patients with drug-resistant epilepsy, surgical intervention becomes a viable option. However, precise localization of the seizure onset zone (SOZ) is crucial for successful surgery. This thesis investigates the potential of the L measure, a non-linear method analyzing directional couplings between brain regions, for SOZ detection in pharmacoresistant epilepsy patients using electroencephalography (EEG) data recorded in a natural environment. We analyzed seizure dynamics in 10 patients using EEG data from the Melbourne NeuroVista Seizure Prediction Trial database. Applying the L measure, we explored connectivity patterns within and across brain regions during pre-ictal, seizure onset, and ictal stages. Network analysis using graph theory metrics assessed these variations across EEG channels and patients to identify potential SOZ locations. Furthermore, we developed a novel method, to track channel connectivity dynamics during seizures, potentially detecting the SOZ with higher temporal resolution. These findings are expected to contribute to a more comprehensive understanding of seizure dynamics and the potential of the L measure for SOZ detection in pharmacoresistant epilepsy patients. This research may pave the way for improved surgical planning and treatment outcomes for this challenging patient population.
How synaptic transmission influences the dynamics of populations of spiking neurons
(2024) Al Hakioui El Mettioui, Zaid
In recent years, models of quadratic integrate-and-fire (QIF) neurons have become ubiquitous in the mathematical neuroscience field. The election of an appropriate synaptic transmission model, either current-based (CUBA) or conductance-based (COBA), is a major challenge that arises from these neuronal frameworks. This problem has been approached using large-scale numerical simulations. However, the lack of use of low-dimensional models obviated the possibility to study how CUBA and COBA approaches shape the dynamics of spiking neurons networks (SNNs) from a general perspective. For this purpose, in this thesis we use a set of exact macroscopic equations for SNNs. This neural mass model (NMM) allows us to perform a comprehensive mathematical analysis of the system’s dynamics, making possible a comparison between the networks’ behavior in terms of the mean membrane voltage and the mean firing rate. Through simulations of different scenarios, including single neuron dynamics and large populations with recurrency, this work uncovers the mechanisms of microscopic and macroscopic state of spiking neurons.The comparison of these approaches advances our understanding of how neuronal signalling is modulated in terms of computational synapses. The results concluded that, unlike the CUBA approach, the COBA model exhibits a non-monotonic and complex dynamics. In this framework, conductance plays a crucial role in shaping the SNNs responses. These findings can be directly correlated with previous research studies where synaptic conductance induced the so-called high-conductance state when tested in vivo.
Characterization of neuronal dynamics in working memory: exploring phase synchronization across spatial, spectral, and temporal dimensions
(2024) Prenafeta Ribau, Joan
Working memory (WM) is an essential neuronal function for higher-order cognitive tasks such as reasoning, language comprehension, problem-solving, and planning. Understanding the mechanisms underlying WM requires examining the communication between brain networks at high spatial, temporal, and spectral resolutions. For this examination, phase synchronization of brain oscillations is often studied in neuroscience to characterize neuronal coordination. Nevertheless, the patterns of synchronization across time, frequency, and space in the context of WM are not fully understood. This thesis aims to provide a comprehensive analysis of phase synchronization of magnetoencephalography (MEG) signals from subjects performing so-called n-back WM tasks. We first use bivariate mean phase coherence to identify synchronized neural subnetworks. Subsequently, we apply the re-normalized mean resultant length to explore the multivariate phase synchrony of subnetworks across frequencies and time. Our findings reveal distinct patterns in nine spectral ranges, spanning from delta to gamma bands. Lower frequencies show greater interhemispheric and long-range connectivity, while higher frequencies emphasize local synchrony within specific brain regions. The analysis is complemented with surrogate signals to confirm that the observed patterns of synchrony are intrinsic to neuronal dynamics. Desynchronization between the sensorimotor cortex and angular gyri suggests the existence of an alpha-mediated attentional control mechanism that inhibits response during the n-back task. In addition, the event-related desynchronization identified in the sensorimotor cortex could indicate the allocation of neural resources to more demanding cognitive processes in memory encoding. Future studies integrating resting-state data may potentially improve our understanding of the neuronal mechanisms underlying WM.
Phase and morphology analysis of cerebral blood flow signals for intracranial pressure characterization
(2024) Pi Mas, Alba
Intracranial pressure (ICP) is an important health indicator for managing patients in neurocritical care. However, since current ICP monitoring methods are invasive, it is only monitored in critical cases. Consequently, there is a need to develop noninvasive monitoring techniques that can broaden the scope of ICP management. This study aims to analyze Cerebral Blood Flow (CBF) signals, which have been shown to undergo morphological changes in response to ICP variations, and identify novel features that hold significance for ICP estimation. We analyzed the morphology and phase evolution characteristics of CBF signals from 20 patients. From these analyses, a total of 6 features were extracted. Linear correlations between these characteristics and ICP were evaluated, and strong correlations of up to 0.9 were found for both analyses. To assess the clinical potential of these features, we conducted a preliminary evaluation of their ability to diagnose high ICP using machine learning. A random forest classifier was trained, and it achieved moderate results. This bachelor thesis highlights the utility of using phase and morphology characteristics of CBF signals as a non-invasive method for ICP estimation, providing a foundation for further studies and potential clinical applications.
Determination of perfusion abnormalities in subjects with status epilepticus
(2023-09-26) Valiente López, Clàudia-Li
Status epilepticus is a condition in which the brain is in a state of seizure activity without returning to its basal state for more than 5 minutes. Medication is the usual treatment, but when it is ineffective, surgical removal of the epileptogenic focus, the area responsible for seizures initiation, is considered. In patients with no apparent brain lesion, perfusion magnetic resonance imaging can be used to identify abnormal blood flow and locate the epileptogenic focus. Several studies have employed perfusion imaging to detect abnormal values in individuals with epilepsy. However, most of these methods need reference values on which to compare and establish what abnormality is. This thesis proposes a methodology to overcome the limitations of the state-of-the-art to assess brain perfusion abnormalities in subjects with status epilepticus. It is based on an individual voxel-based z-score to find outliers, and electroencephalographic recordings to confirm the location detected. A comparison with an asymmetry-based analysis has also been carried out. Additionally, a comparison between dynamic susceptibility contrast (DSC), the gold-standard perfusion imaging technique that uses contrast agent, and arterial spin labeling (ASL), a non-invasive magnetic resonance imaging sequence, has been performed. With the outlier methodology, perfusion abnormalities related to electroencephalographic locations were successfully detected in a similar percentage of patients (90.48% with DSC, 87.5% with ASL) than with the asymmetry-based approach (87.5% with DSC, 100.00% with ASL). In conclusion, the proposed methodology delivers high accuracy and reliability to localize the epileptogenic focus, enhancing presurgical planning for a precise resection of the focus.
Biomechanical regularization in a deep learning network for a fetal MRI registration and segmentation pipeline
(2023-09-26) Recober Martín, Judith
Deformable medical image registration and automatic segmentation are essential procedures in many image analysis applications, such as in early detection and monitoring of fetal and neonatal brain development. Nowadays deep learning can help clinicians in performing these tasks achieving similar or even better results than classical approaches but much faster. V.Comte et al., 2023 presented a novel unsupervised segmentation method based on multi-atlas segmentation, which in fact, avoids the need for a large annotated dataset to train the DL network [1]. This work aims to modify this pipeline by incorporating a biomechanical model to serve as a constraint during CNN training. Biomechanical models that can describe brain deformation can be utilized for DL regularization purposes. Furthermore, this work also tries to access the local shear modulus of the fetal brain, an important measurement that otherwise is nearly inaccessible through other techniques. The best validation Dice score obtained in this work was 0.924 and outperforms some of the state-of-the-art registration pipelines that have been developed, where their validation Dice score was around 0.862 [2] and 0.858 [3]. Moreover, this work was used to obtain local shear and bulk modulus maps of the fetal bran where the cerebellum, the brainstem and the thalamus were found to be the stiffer regions in the brain. Additionally, the results obtained in this project indicate that the white matter exhibits slightly greater stiffness compared to the cortex.
Computational model of the fetal cardiovascular system with aortic coarctation
(2023-09-26) Pellisé Tintoré, Anna
Aortic coarctation (AoC) is a challenging congenital heart disease (CHD) to diagnose prenatally, accounting for 6-8% of all CHDs. It involves the narrowing of a segment of the aortic arch, typically affecting the aortic isthmus. Prenatal diagnosis of AoC is difficult due to the presence of the fetal shunt called ductus arteriosus (DA), which bypasses the defect in fetal circulation. During fetal life, the body adapts to coarctation by establishing mechanisms for adequate oxygen and nutrient delivery as well as pressure regulation to prevent adverse remodelling. However, following birth, a multitude of hemodynamic effects can arise as a result of ductal closure. The primary indicator of AoC is ventricular disproportion, characterized by left-to-right blood flow redistribution resulting in an imbalance with dominant right ventricles. Nonetheless, false-positive diagnoses can occur due to physiological right dominance in the third trimester. The lack of clear prenatal diagnosis criteria for AoC arises from the inconsistency of associated signs. Thus, achieving an accurate prenatal diagnosis of AoC continues to pose a significant challenge. Given the above, this study aimed to develop a closed 0D computational model of the fetal cardiovascular system to improve the understanding of AoC. Particularly, it aimed to simulate the hemodynamic changes in the fetal circulation considering different scenarios of AoC. To that end, we used a closed 0D model that was further extended in order to add more detail and obtain a circuit more consistent with the real anatomical configuration of the fetal system. Then, a thorough parametric analysis was carried out to adjust the parameters of the extended lumped circuit and replicate the behavior of a healthy fetus described in the literature. Finally, multiple scenarios of aortic coaction were simulated considering different degrees of narrowing, ventricular disproportion and increment in the radius of the DA. The obtained results demonstrate the capacity of the extended closed 0D lumped circuit to mimic the hemodynamic behavior of a healthy fetus. Moreover, it has been found that, in the context of AoC, it is imperative for the body to combine the narrowing of aortic segments and the ventricular disproportion, to ensure proper blood delivery, regulate wall stress and wall shear stress in the upper body and most importantly, the brain and the left ventricle. Thus, this study has identified the specific conditions under which physiological ranges are achieved in the context of AoC. Finally, it has been demonstrated that while DA plays a crucial role in blood flow redistribution toward the lower body, it does not experience any increase in its radius.
Classifying consciousness states through an arrow of time inspired framework
(2023-09-26) Patón Baeza, Daniel
Sleep is a natural, recurring state of rest and unconsciousness in which the body and mind undergo a series of physiological and neurological changes. It is divided in 4 stages: REM, N1, N2 and N3, also called deep sleep, which is the lowest level of consciousness that a person undergoes without considering it an abnormal clinical stage. Traditionally, brain activity during sleep has been studied through electroencephalographic recordings. In this work we propose a framework to study N3 using fMRI data only. This work also had the objective of finding a new method to classify different levels of consciousness, given that the actual methods of disorders of consciousness diagnosis usually depend on biased interpretations of the observed reaction of the patient to certain stimuli. The Generative Connectivity of the Arrow of Time (GCAT), a thermodynamics inspired framework, has been used to obtain the generative effective connectivity (GEC) of the brain, which is a measure that includes structural and functional information. In the GCAT framework a wholebrain model is used to determine the causal mechanisms underlying changes in brain hierarchy. Using the GEC, it has been able to classify between wake and deep sleep in human subjects, with an accuracy of over 90%. Furthermore, relevant topological changes have been found between these two states, identifying the brain regions which could be responsible for the loss of consciousness in deep sleep. These results open the door towards expanding the knowledge about sleep stages as well as being the first steps towards achieving a systematic and objective method for disorders of consciousness’ diagnosis.
Longitudinal changes in resting-state network connectivity in youth at familial high-risk for schizophrenia and bipolar disorder
(2023-09-26) Muriel Maillo, María
Psychotic disorders present significant challenges in understanding their causes and developing effective interventions. Functional magnetic resonance imaging (fMRI) has become a valuable tool for investigating these disorders and identifying potential biomarkers. This thesis aimed to explore functional connectivity patterns in restingstate networks (RSNs) among children and adolescent offspring of individuals with Schizophrenia (SzO), bipolar disorder (BpO), and controls. Resting-state fMRI data acquired at baseline and during a follow-up period were analyzed using statistical techniques to examine connectivity between and within RSNs associated with psychosis. The results revealed significant differences in RSN connectivity, particularly in the DMN and CEN. Abnormal functional interactions between the DMN and CEN were observed in the SzO group, indicating an aberrant dynamic between these networks. Age-related variations in connectivity patterns were also found, highlighting distinct associations between RSNs and brain maturation processes in the different groups. The study underscores the potential of fMRI as a tool for identifying objective biomarkers. Moreover, it highlights the significance of including variables that facilitate better extrapolation to the clinical reality of psychiatry. By addressing these considerations, future research can build upon these findings and further advance our understanding of psychotic disorders.
Uncovering seizure connectivity and directionality patterns across patients with focal epilepsy
(2023-09-26) Moreno Creixell, Violant
Epilepsy is a chronic neurological disease characterized by recurring seizures. While most epileptic patients can manage their condition with antiseizure drugs, approximately 25% of individuals experience drug-resistant epilepsy and their only option is to undergo surgery to remove the seizure-generating tissue. However, 65% of these surgeries result unsuccessful. Therefore, this bachelor thesis focuses on exploring a way to identify the seizure onset zone (SOZ) through electroencephalogram (EEG) recordings of seizures. The aim of this study is to analyze EEG seizure data using the Nonlinear Time Series Analysis (NTSA) measure L to identify the connectivity patterns and directionality of signals originating from the seizure onset zone. Our hypothesis is that these signals exhibit distinct connectivity profiles compared to signals from healthy brain tissue. Additionally, the study aims to classify electrodes depending if they are placed in the region which generates the seizures based on their connectivity profiles. The methodology consists of three main stages. Firstly, the measure L is applied to known dynamics like the Lorenz attractor. This measure is a distance rank-based measure able to detect directional coupling from time series that has been proven to be robust to noise and dynamics asymmetries. Secondly, suitable multi-channel EEG datasets with several seizures are searched for. Lastly, the measure L is applied to the EEG recordings to identify connectivity patterns and facilitate machine learning-based electrode classification. The results that are obtained in this thesis show that the connectivity profiles of the seizures evolve in time and allow for this electrode classification. With this, we bring insight into the connectivity dynamics of the SOZ that may allow for a more precise SOZ detection in the future and therefore, improve the resection surgery outcomes.
Computer-aided detection system for pulmonary embolism with integrated cardiac ssessment based on embolic burden
(2023-09-26) Luque Del Toro, Iván
Pulmonary embolism (PE) is a cardiovascular disease caused by one or several occlusions in the pulmonary arteries. Its diagnosis is mainly reliant on imaging, being computerized tomography pulmonary angiogram the gold standard. Recently, there has been increasing interest in automatizing PE detection with the use of computeraided detection systems, aiming to reduce workloads and enhance identification. Semiquantitative scores of embolic burden have also been proposed to characterize PE severity for better patient management. Yet, few attempts have been done to couple both. Here, we propose a system capable of PE detection, which exploits the visual explanations of the detector part to produce 2D representations of embolic burden. These are later used to predict right-to-left ventricle diameter (RV/LV) ratio ≥ 1, a prognosis cardiac feature strongly associated with embolic burden. The detector part is based on a Squeeze-and-Excitation-ResNet50, trained on a subset of the RSNA-STR Pulmonary Embolism CT dataset. The model achieves an accuracy of 0.72, sensitivity of 0.73, and specificity of 0.82 on the test set, which is slightly below the performance of radiologists. As the cardiac assessment module directly depends on the detector’s performance, we were unable to predict RV/LV ratio ≥ 1 successfully. Nevertheless, we believe our system is theoretically feasible and could assist in both PE detection and risk assessment in the future. For that, further work should focus on improving the performance of the detection model, especially regarding high false positive rates, and tune the assessment module accordingly.

Navigate

Browse

Recent Submissions