FCT: DF - Dissertações de Mestrado
URI permanente para esta coleção:
Navegar
Entradas recentes
- Automatização de um porta-amostras para uma câmara de reações nuclearesPublication . Oliveira, Inês Lança de; Fonseca, Maria Micaela; Cruz, JoãoNeste trabalho é desenvolvido um sistema de automatização de um novo porta- amostras para a câmara de reações nucleares do acelerador Tandem de 3 MV do Campus Tecnológico e Nuclear do Instituto Superior Técnico (CTN/IST). Neste sistema, o novo porta-amostras, com capacidade para 12 alvos igualmente distribuídos por 3 faces, é mo- vimentado verticalmente até 50.8 mm com recurso a um linear feedthrough e em 360º com uma plataforma rotacional. Estes componentes são automatizados com recurso a dois motores de passo comandados por um arduino. Uma interface desenvolvida em Python permite ao utilizador controlar a movimentação do porta-amostras remotamente com uma precisão estimada de 0.1 mm na vertical e 0.3º angularmente. Adicionalmente, construiu-se e montou-se na câmara uma gaiola de Faraday para o novo porta-amostras que permitirá coletar os eletrões secundários emitidos pelo alvo. O sistema desenvolvido será testado na câmara de reações nucleares posteriormente a este projeto devido à necessidade de se fabricar uma nova tampa para a mesma com o design adequado para o novo sistema de automatização. Contudo, é proposto um conjunto de amostras que permitirão efetuar testes de carga à nova gaiola de Faraday e verificar a reprodutibilidade de resultados do novo sistema. Foi feito um conjunto de simulações com os programas SIMNRA e ERYA-Bulk tendo em conta as condições experimentais atuais na câmara de reações nucleares. Os resultados esperados são apresentados para uso futuro na testagem da câmara de reações nucleares melhorada. Os laboratórios de aceleradores de partículas de pequenas e médias dimensões, como o CTN/IST, são espaços de excelência para a aplicação de técnicas analíticas de feixe de iões (técnicas IBA), requisitadas pelas mais diversas áreas do conhecimento como medicina, conservação e restauro, astrofísica, caraterização de materiais, entre outras. A eficiência dos equipamentos disponibilizados aos investigadores é determinante para a disponibilidade de tempo de feixe. Neste contexto, a melhoria das linhas dos aceleradores e a automatização do processo experimental são exploradas sempre que possível.
- REDUÇÃO DA ATIVIDADE ADMINISTRADA ATRAVÉS DA MELHORIA DA QUALIDADE DA IMAGEM EM CINTIGRAFIA PEDIÁTRICAPublication . Arsénio, Marta Sofia Rocha; Mota, Ana; Vigário, RicardoO objetivo deste estudo foi reduzir a atividade administrada em imagens de cintigrafia renal pediátrica por meio da aplicação de técnicas de melhoria de qualidade da mesma. As imagens obtidas após a administração do radiofármaco 99mTc-mercaptoacetiltriglicina (99mTc-MAG3) foram submetidas a diferentes processos de redução de ruído. A redução da atividade administrada foi simulada através da soma parcial dos dados das imagens originais (100 %, 75 %, 50 %, e 25 %). Para melhorar a qualidade das imagens, foram aplicadas quatro redes neuronais (DnCNN, UDnCNN, DUDnCNN, e AttnGAN) e um Filtro de minimização da variação total (TVmin) aos dados. A qualidade das imagens foi avaliada através da Razão Sinal- Ruído (SNR) dos rins e do Índice de Similaridade Estrutural Multiescala (MS-SSIM), comparando a preservação dos detalhes com a redução do ruído. Imagens simuladas com menor atividade administrada (25 % dos dados) apresentam níveis mais altos de ruído. O uso do filtro TVmin e da rede UDnCNN resultou em melhorias significativas na qualidade das imagens. A rede UDnCNN destacou-se pelo seu equilíbrio eficaz entre a redução de ruído e a preservação dos detalhes estruturais. Além disso, a combinação do filtro TVmin com a rede UDnCNN demonstrou ser promissora devido ao seu tempo de teste mais rápido. A análise dos tempos de treino e teste revelou que a combinação dessas técnicas oferece um bom equilíbrio entre a qualidade da imagem e a eficiência de processamento. A abordagem integrada pode permitir a aquisição de imagens com atividades administradas reduzidas (cerca de 50 %) sem comprometer a qualidade, representando um avanço significativo na redução de radiação administrada aos pacientes. Este estudo demonstra que a integração de técnicas avançadas de processamento de imagem e redes neuronais pode melhorar significativamente a qualidade das imagens de cintigrafia renal, permitindo a aquisição de imagens de alta qualidade com atividades administradas menores de radiação, beneficiando a população pediátrica.
- Estudo dos sistemas simpático e parassimpático através da modelação de dados de pupilometriaPublication . Silva, Bernardo José Gervásio Canelas da; Quintão, Carla; Vieira, PedroRecebendo inervação do Sistema Nervoso Autónomo Simpático (SNAS) e do Sistema Nervoso Autónomo Parassimpático (SNAP), a abertura da íris, a pupila, é regulada por estes dois sistemas e, por isso, fornece informações que poderão ser relevantes do ponto de vista clínico [1]. Nomeadamente, permite obter informações sobre o estado de consciência de um indivíduo, ou servir como diagnóstico de certas doenças neurodegenerativas, como são os casos das doenças de Parkinson ou de Alzheimer [2], [3]. Tendo em conta a relevância clínica dos sistemas simpático e parassimpático, pretende-se primeiramente estudar as frequências do seu normal funcionamento, com o intuito de vir a compreender quais as alterações observadas em situações patológicas. Neste trabalho, estudou-se a influência de um estímulo frio sobre as dinâmicas da pupila e do coração, numa amostra de 10 indivíduos. Pela literatura [4], este tipo de estímulos evidencia a atividade simpática sendo, pois, possível separar a ação dos dois sistemas, que, muitas vezes, atuam em simultâneo. Utilizou-se um pupilómetro de alta resolução para adquirir os dados de pupilometria, e estudou-se a variabilidade da frequência cardíaca (VFC) dos indivíduos, com base no seu espectrograma. À semelhança do que acontece na análise do eletrocardiograma (ECG), supõe-se que possa haver uma frequência própria do funcionamento da pupila. Adquiriu-se o sinal que representa a variação da área da pupila ao longo do tempo. O cálculo da energia das componentes extraídas desse sinal com a ferramenta matemática Singular Spectrum Analysis, análise de espectro singular (SSA) permitiu concluir acerca da atividade do Sistema Nervoso Autónomo (SNA). Os resultados embora pouco conclusivos no que respeita à banda de frequências que caracteriza o normal funcionamento do SNA, com base nos dados da pupilometria, permitem concluir que existe um aumento na energia das altas frequências depois da resposta ao estímulo frio, o que vai ao encontro do que se esperava obter, tendo em conta a bibliografia referente à atividade cardíaca.
- Enhancing Epilepsy Surgery: Generating Synthetic Multi-channel iEEG Data with GANs for Improved Tissue IdentificationPublication . Marques, Beatriz dos Santos; Gamboa, Hugo; Silva, LuísIn epileptic surgery, the scarcity of available invasive electroencephalogram data impairs the development of accurate classification models able to distinguish between healthy and injured brain tissue. Synthetic data has been proposed as a solution. However, the majority of research has focused on synthetic single-channel EEG signals, with a lack of consensus on the quality metrics employed, as the majority focus on the performance improvement of classifiers. The present work proposes a novel hybrid architecture capable of generating synthetic multi-channel intra-operative electrocorticogram (ioECoG) signals. Its backbone is a Deep Convolutional Wasserstein Generative Adversarial Network with Gradient Penalty (DCwGAN-GP). The hierarchical Transformer-based critic extracts features, across time and space, from local segmented windows of the input, and captures the existing rela- tionships across the windows on a global level, reflecting the natural spatio-temporal dependencies of brain activity in the generated signals. The quality of the synthetic data is first audited at a sparse level, on the domains of Fidelity, Diversity and Privacy. The results indicate that existing dependencies of real data, and their variability, were captured, confirmed by a coverage score of 0.919. Some disparities in signal dynamics were observed, suggesting some difficulty in modelling long-term activity and individual patterns. The existing spatio-temporal inter-channel behaviours were evaluated by two new proposed metrics, demonstrating the similarity between the two distributions, confirming the preservation of the natural dependencies. This study generated realistic synthetic multi-channel ioECoG signals displaying interconnectivity patterns of real data, a stepping stone in the development of reliable synthetic datasets for the improvement of robust classifier models that can accurately identify epileptogenic tissue.
- Follow-up process automation based on machine learning. An approach to monitoring post-discharge patientsPublication . Santos, Matilde Martinho dos; Lapão, LuísHospital readmissions pose a significant challenge to healthcare systems, leading to both financial strain and negative effects on patient outcomes. Post-discharge monitoring, particularly through follow-up calls, is crucial in identifying complications early and preventing unnecessary readmissions. However, the existing follow-up processes are often inefficient and resource-intensive. This dissertation aims to analyse the current follow-up process at a Private Hospital in Portugal and develop an artefact that enhances post-discharge care by implementing an ML-driven risk stratification model. The proposed solution automates the follow-up process, classifying patients into low-, moderate-, and high-risk categories based on their post-discharge responses. The system reduces the number of follow-up calls required, optimising nurses’ workload while maintaining care quality. Integrating a machine learning model to assess risk in real-time using self-reported data represents an innovative step in digital health. Three machine learning models were tested: XGBoost, Random Forest, and Logistic Regression. XGBoost excelled in maximising recall and F1 score, while Logistic Regression achieved a higher AUROC, demonstrating its superior ability to distinguish between classes. After analysing the confusion matrices, Logistic Regression was selected as the best model due to its balanced performance across key metrics and its capacity to minimise critical errors. Results indicate that the system can reduce the workload by up to 79%, allowing nurses to focus on high-risk patients and improving efficiency in patient management. Additionally, extending follow-up to 30 days has the potential to prevent avoidable read- missions. The clinical team evaluated the proposed system, confirming its practicality and alignment with healthcare needs. This research highlights the potential of AI and digital tools in enhancing post-discharge monitoring, ultimately improving patient outcomes and healthcare resource allocation.
- Electrical stimulation prototype device for a three-dimensional organ-on-a-chip trainingPublication . Moniz, Beatriz Vargas; Igreja, Rui; Almeida, HenriqueCardiovascular diseases are a leading cause of mortality, driving the need for advanced tools in cardiac tissue engineering. Electrical stimulation is a key technique for promoting cardiac tissue maturation, but existing devices are often expensive, complex, or lack versatility. This research presents the design, fabrication, and validation of a low-cost, highly customizable electrical stimulation prototype for organ-on-a-chip applications. The device, centered on an Arduino Nano microcontroller and a custom Printed Circuit Board (PCB), integrates an H-bridge for biphasic waveform generation, a digitally-controlled amplitude regulator, and fully adjustable temporal parameters, including pulse width and inter-pulse interval. Hardware validation confirmed exceptional performance, demonstrating high accu- racy with minimal relative errors in voltage output 0.99% at 7V, inter-pulse interval timing 0.00% at several intervals, and pulse width control as low as 0.02% at 60ms.Long-term stability tests further affirmed the system’s reliability. Preliminary functional valida- tion using C2C12 myoblasts demonstrated biological effectiveness. Electrical stimulation significantly promoted myogenesis, resulting in larger, aligned myotubes compared to controls. This work successfully delivers an accessible, precise, and reproducible platform to advance cardiac tissue research and therapeutic development.
- AGE DEPENDENCE IN CORTICOMUSCULAR CONTROL - A SYNCHRONY AND SERIOUS GAMING APPROACHPublication . Silvestre, Joana Patrícia Prates; Vigário, Ricardo; Pereira, CarlaUnderstanding how information is transmitted through the Nervous System (NS) is one of the greatest challenges in Neurophysiology. Corticomuscular communication plays a central role in motor control and can be quantified through the Phase-Locking Value (PLV), which measures the synchronization between cortical and muscular activities. Although corticomuscular coupling has been widely studied in adults, developmental differences between different age groups remain weakly understood. Evidence suggests that the relation linking brain and muscular activities can be revealed through synchrony of electrical signals measured at the scalp - EEG (Electroencephalography) - and the muscle - EMG (Electromyography). This dissertation investigated age-related differences in corticomuscular synchrony during a set of motor tasks with different levels of difficulty. EEG and EMG signals were recorded from healthy adults and children while performing those tasks. After preprocessing, Independent Component Analysis (ICA), Temporal Decorrelation Source Separation (TDSEP) and Reference Phase Analysis (RPA) were applied to isolate neuronal sources, followed by source localization of cortical generators involved in motor control. Within-group analyses contrasted dominant vs. non-dominant hand, unilateral vs. bilateral performance, maximal-force vs. all other tasks, and – only in adults – task complexity. Between-group comparisons were conducted task by task. Results showed consistently higher PLV in adults than in children, indicating stronger corticomuscular synchrony with maturation. In both groups, dominant-hand actions yielded higher PLV, and the maximal-force task elicited the highest PLV. Source localization revealed distinct patterns: adults showed consistent activations focused in the primary motor cortex (M1) and adjacent motor areas, while children exhibited more diffuse and heterogeneous activations, reflecting less mature corticomuscular control. These findings indicate that corticomuscular synchrony is age-dependent and varies with task characteristics, particularly hand dominance and maximum force contractions. The results contribute to a better understanding of motor control across the lifespan and may inform future applications in rehabilitation and assistive technologies.
- Personalized dosimetry in radionuclide therapy: SPECT simulations with Monte Carlo-SIMINDPublication . Dias, Maria Luísa de Borba Aguiar; Ferreira, Paulo; Cruz, JoãoSingle Photon Emission Computed Tomography (SPECT) images of 177Lu-radioligand therapy (RLT) are of paramount importance for 177Lu-RLT dosimetry. Digital SPECT scanners can help determine the best parameters to be used in the real SPECT acquisitions, and consequently improve dosimetry estimation. Ultimately, this work aims to evaluate SIMIND’s ability to accurately reproduce images of patients treated with 177Lu. SIMIND Monte Carlo software (v8.0) was used to simulate projections to be recon- structed with the PyTomography library, based on the OS-EM algorithm. The SIMIND SPECT scanner was validated by comparing the simulation results of the NEMA NU 1-2007 tests with measurements performed on the Philips BrightView SPECT scanner. Intrinsic spatial resolution, linearity, intrinsic energy resolution and system planar and volume sensitivity were tested. The first three tests employed 99𝑚 Tc, while the sensitivity experiments also included 177Lu. Recovery Coefficients were also tested. A SPECT scan of the NEMA IEC Body Phantom was acquired in the real and simulated scanners with 99𝑚 Tc and 177Lu. Real and simulated SPECT projections from two patients were reconstructed for three available time points. Mean absorbed dose in the VOIs was computed through Dose Voxel Kernel (DVK) convolution with the obtained real and simulated TIA maps. The validation results were satisfactory with errors below 10%, apart from the in- trinsic spatial linearity. Recovery coefficients determined with NEMA phantom showed good agreement between real and simulated, with higher disparities observed for 177Lu compared to 99𝑚 Tc. Obtained mean absorbed doses were overall consistent for real and simulated patients. Partial Volume effects could be responsible for the differences observed in the kidneys. Tumours presented the highest differences, which could be explained by their size. SIMIND and PyTomography showed potential to allow personalized dosimetry in 177Lu-RTL patients.
- Neural Entrainment to Prosodic Structure: a phase-locking approachPublication . Vicente, António Guilherme dos Santos; Vigário, Ricardo; Frota, SóniaAs technology and scientific inquiry progress, previously unanswered questions are eluci- dated, while new enigmas emerge. The field of neuronal research concerning the tracking of speech's linguistic structure exemplifies this paradigm. Numerous investigations have demonstrated and accentuated a correlation between speech stimuli and language structure, which induce neural entrainment across various fre- quency bands. These oscillations are predominantly associated with theta and delta neural rhythms. However, there is a scarcity of studies examining the coupling of prosodic aspects of the speech signal and the neural signal at a more profound level within the temporal domain. This can be accomplished through studying specifically the phase domain of the signal under scrutiny. This thesis seeks to investigate neural entraining between speech stimuli, presented under two distinct prosodic conditions, and their corresponding EEG response. That may be useful to improve our understanding of the prosodic structure of speech. To accomplish this, the Reference Phase Analysis algorithm was employed to identify the neural sources exhibit- ing the strongest phase synchronization with the speech stimulus. We observed that, in particular for frequencies associated with syllable and intonation, separated signal sources exhibit higher phase-locking values than any recording EEG channel data. Also, corroborating that finding, we observed that most of the found sources presented a rather focal and dipolar distribution of weights across the EEG cap, suggesting neuronal locations in parr with what can be found in literature for said stimuli.
- Automated Microfluidic Platform for Quantitative Analysis of Molecular Trans- port across Biomimetic InterfacesPublication . Pais, Beatriz dos Reis Antunes; Neves, Miguel; Cruz, JoãoA comprehensive understanding of human physiology and disease is crucial for ad- vancing drug discovery and developing effective therapies. This requires physiologically relevant in vitro models supported by standardized, robust, and reproducible methods. Traditional 2D cell cultures and animal models often fail to mimic the dynamic microenvi- ronments and selective transport functions of human tissue barriers. While microfluidic barrier models offer better biomimicry, many are limited by technical complexity, lack of standardization, and insufficient real-time analysis capabilities, especially under non- equilibrium conditions. Non-equilibrium molecular transport is a critical feature of many biological barriers, including the renal tubule, blood–brain barrier, intestinal epithelium, and placental interface. These systems rely on active transport mechanisms, directional fluid flow, and dynamic concentration gradients that cannot be accurately modeled us- ing static or oversimplified systems. This work presents a fully automated microfluidic platform with programmable control and closed-loop feedback for standardized analysis of molecular transport across biomimetic tissue barriers. It enables dynamic regulation, precise fluid handling, and high-resolution monitoring, with continuous compensation for solvent drag and hydrostatic pressure. Proof-of-concept validation demonstrates the system’s potential to minimize user intervention, improve reproducibility, and support inter-laboratory standardization. The platform offers scalable, high-fidelity modeling of barrier function under dynamic conditions, with broad applications in translational research and preclinical drug development.