APRÈN. Portal de la producció docent del professorat de la UPC

This research investigates the influence of bio-inspired Region-of-Interest (ROI) prior information on low-power event-based Spiking Neural Network (SNN) eye tracking systems. The proliferation of edge computing devices has catalyzed significant advancements in neuromorphic engineering, particularly within the domain of vision processing systems due to the need for in-sensor or near-sensor operations. Neuromorphic vision systems, leveraging event-based sensors and spiking neural networks, demonstrate substantial potential for wearable eye tracking applications where power consumption and computational footprint require optimization. These systems generate asynchronous output only upon detecting local brightness changes, thereby significantly reducing data redundancy and power consumption while maintaining temporal precision. The present study addresses the performance-efficiency trade-off in neuromorphic eye tracking by developing and evaluating a novel processing pipeline that incorporates multiple bioinspired mechanisms to efficiently determine regions of interest within event-based visual data. The proposed methodology implements a four-stage processing sequence: spatiotemporal sustained polarity filtering emulating retinal persistence, spatial orientation selectivity through Linear Discriminant Analysis, center-surround excitation/inhibition filtering for edge enhancement, and winner-take-all attention for ROI vertex selection. This bio-inspired approach was integrated with the neuromorphic Retina network architecture (64k parameters) and systematically evaluated on the 3ET+ dataset across diverse recording conditions. Experimental results demonstrate that the ROI-enhanced configuration achieves a 14.8% reduction in mean localization error and significantly improved predictive consistency (+ 18.6 percentage points) compared to the baseline. When implemented with selective prediction through a one-class detection approach filtering out wrongly determined ROI crops from training and evaluation, the system demonstrates a 51.5% reduction in localization error (3.529 vs. 7.281 pixels) and a 45 percentage point improvement in accuracy metrics (88.2% vs. 43.2%), hence hinting at potential attainable gains by keeping on improving the ROI determination. Concurrently, the ROI-enhanced architecture exhibits 30.1% lower firing rates and a 9.5% reduction in synaptic operations, enhancing computational efficiency. This research advances neuromorphic vision processing through the implementation of bioinspired spatial attention mechanisms within hardware-constrained processing pipelines. The methodological approach demonstrates how computationally efficient feature extraction from event-based visual data can be achieved through selective processing, thereby optimizing the allocation of limited neuromorphic computational resources. These findings have significant implications for ultra-low-power eye tracking systems in resource-constrained applications, providing a framework for enhancing tracking precision while maintaining compatibility with advanced neuromorphic hardware platforms such as Speck, Loihi, and SpiNNaker.

Motor impairments resulting from neurological disorders such as stroke often lead to significant disability, requiring intensive rehabilitation. Depending on the severity of the brain lesion, some patients may not qualify for conventional physical therapy due to limited residual motor function. Brain-Computer Interfaces (BCIs), particularly those based on Motor Imagery (MI), offer a promising alternative for supporting neurorehabilitation in severely impaired individuals. MI has been proven to elicit brain activation patterns similar to actual movement, making it especially useful in cases where overt motor execution is not possible. This project aims at designing and developing a functional framework for an MI-based BCI capable of classifying left versus right hand motor imagery and delivering real-time neurofeedback. EEG signals were acquired from eleven healthy subjects during a standardized MI task. The data were used to implement an offline processing pipeline that included signal preprocessing, feature extraction using Common Spatial Patterns (CSP) algorithm, dimensionality reduction through feature selection, and classification using various supervised learning algorithms. Classification performance varied considerably across individuals, with 7 out 11 subjects achieving accuracies above 80%, and the best subject reaching 91.6%, while others scores close to chance level. These results confirm both the feasibility of the proposed system and the strong inter-subject variability often observed in MI-BCI applications. An online prototype was also developed, integrating real-time EEG acquisition, CSPbased feature extraction, classification, and visual feedback. Although online testing was limited to a small number of sessions due to time constraints, the system demonstrated its capacity to detect and respond to MI tasks in real time. Two additional BCI applications based on blink detection were implemented as demonstrators, showing high responsiveness and positive user feedback in public science outreach events. Despite promising results, the system's generalization remains limited, and the lack of data from motor-impaired individuals represents a noteworthy constraint. Future work should focus on improving the robustness and adaptability of the real-time system through extended calibration, adaptive learning, and testing with clinical populations. With continued development and validation, the proposed MI-BCI framework holds potential as a cost-effective and accessible tool for neurorehabilitation.

Epilepsy, a neurological disorder affecting millions worldwide, presents significant challenges in its drug-resistant form, where standard treatments often fail. The GALVANI project aims to address these challenges by developing personalized whole-brain models to optimize therapeutic strategies such as transcranial current stimulation and surgery. These models integrate structural, functional, and clinical data to simulate seizure dynamics and predict treatment outcomes. This thesis focuses on advancing the pipeline for whole-brain modeling by transitioning from invasive stereo-electroencephalography (SEEG) to non-invasive scalp electroencephalography (EEG). Using neural mass models (NMMs) and structural connectivity derived from diffusion MRI, the study develops a framework to personalize whole-brain models using EEG data. Key steps include generating synthetic EEG via forward modeling, comparing it to empirical EEG, and optimizing model parameters such as excitability. Validation results highlight the feasibility of capturing seizure propagation patterns using EEGbased models. In particular, topographic amplitude patterns (topographic maps) show a strong spatial correspondence between synthetic and real EEG, especially in regions identified clinically as seizure onset zones. Although functional connectivity comparisons show some qualitative similarities, further work is needed to improve their quantitative alignment. Frequency-specific analyses underscore the relevance of tailoring the model to epileptically meaningful bands, such as theta and gamma, which appear most sensitive to ictal activity. By reducing the invasiveness of data acquisition while maintaining biologically plausible seizure propagation dynamics, this work contributes to the development of clinically applicable, patientspecific simulations for epilepsy treatment. Future directions include integrating multimodal data and extending the framework to other neurological disorders.

El presente Trabajo de Fin de Máster (TFM) se centra en el análisis de señales electroencefalográficas (EEG) para detectar indicios de conciencia en pacientes diagnosticados con estados alterados de conciencia (EAC), como el síndrome de vigilia sin respuesta (SVSR) o el estado de mínima conciencia (EMC). A través de un protocolo basado en tareas de imaginería motora, se ha buscado identificar patrones neurofisiológicos en EEG que evidencien intenciones cognitivas en ausencia de movimiento voluntario observable. La muestra incluyó 19 pacientes con daño cerebral adquirido (DCA), a quienes se aplicó un protocolo específico con mediciones EEG para evaluar diferencias entre condiciones de descanso e imaginería motora. Se observaron respuestas cerebrales consistentes en algunos participantes, lo cual refuerza la necesidad de mejorar los criterios diagnósticos tradicionales. Estos hallazgos podrían contribuir a una transformación en la evaluación clínica de la conciencia, ofreciendo alternativas más precisas y éticamente responsables en el abordaje de pacientes no comunicativos.

Brain-computer interfaces (BCI) have been studied for decades for their potential to control devices by monitoring brain activity. Specifically, MI-based BCI systems (Motor Imagery) have proven highly effective in the field of neurorehabilitation, as the brain patterns generated closely resemble those produced during actual movements. This work proposes a protocol designed to acquire and analyze BCI data using the Bitbrain Hero Helmet device, aiming to explore its use in practical applications. The project is part of POSMOFYA, an acronym for the Hybrid Platform of OrthosisWheelchair to ensure compatibility of Mobility, Functionality, and Acceptability for use in domestic environments. This hybrid platform, as its name suggests, integrates an automated wheelchair and a robotic arm. The ultimate goal is to enable efficient and practical control in domestic environments. The developed protocol is based on principles of robustness and precision, aiming to enhance the ability to recognize users' movement intentions. This work is structured into several phases, beginning with the experimental design. Using the PsychoPy software and the LabStreaming Layer (LSL) protocol, motor imagery tasks were defined focusing on two specific actions: right wrist flexion and left wrist flexion. These tasks were designed to generate brain patterns associated with clear movement intentions, with the goal of training a Machine Learning (ML) system. The strategy ensures precise and consistent capture of brain signals, essential for advancing the development of practical applications within the project framework. The results reflect the challenges of working with EEG signals. Although the Decision Tree algorithm applied to Common Spatial Patterns (CSP) achieved a validation accuracy of 42.9% in offline mode, and K-Nearest Neighbors (KNN) reached 51.0% accuracy with literature-based features, the predictions were biased towards a single class. This highlights limitations in signal discrimination and variability between sessions. Despite the difficulties, the project has provided valuable lessons: the need to improve EEG systems and explore advanced techniques such as Regularized CSP or hybrid methods with deep learning to increase robustness and precision. Additionally, the basic steps of the online pipeline were validated, indicating that better optimization could result in more reliable practical applications. These findings not only highlight the field's challenges but also point to opportunities for improvement in future research.

Despite significant advances in neuroscience, there is still a considerable gap in knowledge about brain connectivity. The complex network of neural pathways that facilitate communication between different brain regions is not yet fully mapped or understood. This lack of comprehensive knowledge forms the basis of this thesis, which aims to develop brain connectivity models. By building these models, this thesis seeks to improve our understanding of how the brain is interconnected, based on the extensive but incomplete research that has been done over the years. The conception of this thesis aims to help advance a project carried out by Neuroelectrics, a company that seeks to develop non-invasive stimulation treatments to combat epilepsy, depression, and Alzheimer's, among others. The research carried out in this work has improved the brain model over a previous one. This enhancement is attributed not only to the extensive literature review but also to the specific and deterministic experiments conducted to estimate the synaptic weight values. These values, which were previously arbitrary, are now precisely determined, leading to a more accurate and reliable brain model and preserving critical brain dynamics.

New technologies are revolutionizing the field of motor rehabilitation. Thanks to image estimation systems, it is now possible to accurately track patient movements, providing additional information to that already available, acquired through neuromuscular signals. Therefore, the aim of this work has been the development of a tool that allows the simultaneous acquisition of the patient's body position and neuromuscular signals. The methodology followed includes the use of a bioelectric amplifier for the acquisition of neuromuscular signals, an Intel Real-Sense camera for capturing body position, and a signal generator for synchronization. The application consists of an interactive graphical interface that displays the amplified signals in real-time and delivers the recorded and synchronized information for post-processing. The results obtained show that the tool is capable of accurately capturing and synchronizing neuromuscular signals and body position, also providing an example of postprocessing both signals. In conclusion, the developed tool has great potential to improve records in rehabilitation, being organized to anticipate future implementations that expand multiple functionalities, such as the integration of more types of amplifiers and the development of more advanced signal processing algorithms.

Right heart complications are particularly common among heart transplant patients, where the right ventricle (RV) faces unique challenges in adapting to the new hemodynamic environment. Mathematical models of the cardiovascular have been developed to estimate hemodynamic states from known properties of the RV such as end-systolic elastance of the RV (Ees RV) and diastolic stiffness constant of the RV (Brv). These models effectively address the forward problem by predicting hemodynamic states based on these parameters. However, the inverse problem, which entails deducing the underlying hemodynamic parameters from clinical measurements, remains unsolved. The primary objective of this project is to develop a system for identifying and solving the inverse problem of predicting hemodynamic parameters, particularly RV parameters, from clinical measurements obtained by right heart catheterization. Validation of these estimates included the creation of a model to predict clinical outcomes and validate the accuracy of the hemodynamic parameters used as predictors. This study employed a k-Nearest Neighbors (k-NN) algorithm for System Identification for Hemodynamic Parameters (SIHP), implemented using Python and Milvus for data management. The algorithm was evaluated on simulated data to predict hemodynamic parameters by identifying the closest states in the dataset and averaging the nearest neighbors' parameters. It is validated the algorithm's parameter estimates using three criteria: by demonstrating its accuracy on simulated data, by showing faithful state reconstructions on patient data, and by using the estimated hemodynamic parameters to successfully fit models predicting patient length of stay (LOS) in the intensive care unit (ICU). It is thus demonstrated as a practical tool that can be applied to real-world patient data to quantify RV function.

This project deals with the design and implementation of a framework capable of generating hand motor decoding models from electromyographic (EMG) signals. The intention is to use these decoding models in the control of hand prostheses. For this purpose, the development has been oriented based on the paradigm of Learn from Demonstration (LfD), which is based on learning complex tasks through observation and imitation of human behaviour. The idea was to develop a system that a person with hand amputation could use to train his or her prosthesis to perform specific movements. Furthermore, this system adapts to the characteristics of the user, being independent of the muscles selected for control or the movements to be performed. With this, the system consists of two phases: a training phase in which the subject shows the movements to be performed by the prosthesis with their healthy arm while generating the control signals in their affected arm, and an evaluation phase in which the result of the decoding is compared with the expected movements. To do this, a stereo vision system was used in conjunction with markerless hand kinematics detection software. EMG signal acquisition equipment was used. A robotic hand has been built and programmed to serve as a prosthesis to be controlled in the system. Software based on a ROS environment has been developed for data management and communication. And two models have been trained for decoding: a multiple linear regression model and a Linear Discriminant Analysis (LDA) classifier. After the implementation of the system, an experimental protocol has been carried out in order to evaluate the performance of the system. In this protocol, a group of 6 volunteers without amputation have carried out tests that have allowed the validation of the system, as well as analysing the performance of the different parts that compose it. With this, it has been verified that the system is capable of generating the models proposed. It has also been seen that the two models used in this work are capable of identifying the user's movement intentions. In conclusion, the implemented system successfully meets the proposed objectives. And thanks to the modularity offered by ROS, it opens the door to future developments that integrate new models for decoding or use more advanced commercial prostheses into the system.

This project centers around the integration of electromyography (EMG) and gaming technology to design a tool aimed for motor rehabilitation practices. The design includes a user-friendly interface and a serious game controlled by muscle contractions, providing immediate biofeedback and making rehabilitation engaging. The methodology hinges on using a series of Python libraries to develop a game based on the classic video game Flappy Bird, where the bird's motion is controlled by EMG signals reflecting muscle contractions. This approach ensures the game is both accessible and effective in meeting diverse rehabilitation needs. The tool not only extracts physiological indices from the EMG signal for clinical analysis but also incorporates a dashboard that presents these metrics to healthcare professionals, enhancing medical oversight and analysis. The rehabilitation tool successfully implements four therapeutic modes, each tailored to different rehabilitation scenarios. This setup demonstrates the tool's versatility in adapting to various user requirements for effectively enhancing motor function rehabilitation. Conclusively, the tool successfully meets the expected outcome by integrating EMG signal analysis with gamification, providing effective biofeedback-driven rehabilitation and EMG indices for medical analysis. Future work will concentrate on refining the dashboard based on professional feedback, expanding game modalities, and enhancing the rehabilitation experience with additional visual and audio elements.

This thesis undertakes a thorough investigation on the Density Spectral Array (DSA) parameters in Depth of Anesthesia (DoA) monitoring devices, emphasizing its application on the CONOX® (Quantium Medical S.L) monitoring device, by considering its own particularities. The aim is to develop a personalized solution for the optimization of the spectrogram visualization of the electroencephalogram (EEG) signal, considering patient inter-variabilities, to finally provide a useful tool for the assessment of the anesthesia practice and Loss of Consciousness (LOC) on a surgical environment. The followed methodology includes an analysis of 1B, 2B and 3B byte depths and several alpha values in exponential averaging, to determine the best fit for the EEG data and the company's interests, which resulted in the selection of 1 byte depth and an averaging alpha of 0.2. Furthermore, 11 equations are postulated and analyzed to improve the rescaling of the raw EEG, providing an age-dependent solution for four age ranges covering from 16 years old to 93, which facilitates an optimal visualization of the spectrogram, by selecting three equations for each age range; a default equation for majority of cases, a red saturating equation for low EEG energy patients, and a blue saturating equation for high EEG energy patient, allowing to pivot between equations during real time when required.

The main objective of this project is to develop a chatbotspecialized in answering questions about renal lithiasis with clinically validated information using technologies in full expansion such as natural language processing and machine learning. These technologies allow the chatbot to understand and respond intelligently to the queries and questions of the users, offering personalized and customized answers to each case. The methodology established for the project consists of three main blocks. The first is based on gathering information from reliable medical sources to create a solid knowledge base on which to base the chatbot's answers. The second block consists of conducting a systematic search of the different tools available to create a chatbot and stipulating the selection criteria to determine which one most efficiently fits the needs of this project. Finally, in the third major block, the process of developing the chatbot using the Botpress platform and the established validation methods by which extensive testing was performed to verify the chatbot's functionality is described. Although the chatbot is considered to have achieved the main objective of the project with a satisfactory level of functionality and accuracy in the delivery of information on renal lithiasis, it is recognized that the tool has the capacity for continuous improvement by adding new functionalities, updating the knowledge base with the latest medical information, and improving the conversational interaction with users. In conclusion, the project has succeeded in developing a chatbot specialized in renal lithiasis that provides clinically validated information in a fast and accessible way. The chatbot offers advantages both for patients, by providing immediate answers to their questions, and for specialists, by optimizing their time and resources. The importance of continuous improvement and user feedback is recognized in order to continue to evolve and provide an increasingly effective service in the field of renal lithiasis.

This thesis involves the design, construction and programming of a device capable of identifying LED sources using a high-resolution camera and an image processing system. Initially, it was considered using a digital microscope, which was cheaper than the high-resolution camera that was used in the end. Due to the problems it had because of its resolution, finally it was opted for using a high-resolution camera. A custom housing was designed using AutoCAD software and built with a 3D printer for preventing sample light contamination. An image processing algorithm was programmed using Python software. It was very useful for identifying the pixels on the image created by the spectrums of the light emitted by LEDs. With those samples, a database was created. The software was capable of identifying which type of LED it was by using it. In the end, the software achieved a 95% accuracy score. All the results were collected and analyzed in order to draw conclusions about the prototype's performance.

Els models d'hospitalització psiquiàtrica tradicionals han estat criticats per ser restrictius i vulnerar drets fonamentals, cosa que pot ser contraproduent per a la recuperació dels pacients a llarg termini. En resposta a aquesta situació, s'han desenvolupat nous models d'hospitalització que pretenen crear un ambient terapèutic segur i que promogui la recuperació del pacient, minimitzant la necessitat de mesures coercitives. L'Hospital Germans Trias i Pujol, ha dissenyat un model dual, que combina un mètode assistencial, anomenat Safewards, amb una eina de predicció de riscos a urgències i una categorització de riscos i de càrrega assistencial a planta a través d'una puntuació global. L'objectiu del model és obrir les portes de la unitat de manera segura, tenint en compte els riscos i el nivell de treball de l'equip assistencial. Les eines que es fan servir actualment no estan validades científicament, dificultant així l'escalabilitat del model. La finalitat d'aquest projecte és elaborar una proposta d'intervenció per desenvolupar dos algorismes d'Intel·ligència Artificial, amb variables validades, que prediguin l'ambient de la unitat i el risc d'un pacient a l'ingrés. En primer lloc, s'ha fet un estudi qualitatiu amb personal expert de la unitat de psiquiatria de l'HGTiP per objectivar els criteris pels quals es defineix el nivell de risc i la càrrega assistencial a planta. Se n'han extret un total de 35 ítems i s'ha desenvolupat una plataforma digital per emmagatzemar-los. Aquests seran utilitzats, juntament amb altres variables dinàmiques, com a entrada en un algorisme de regressió. D'altra banda, es desenvoluparan algorismes de classificació per predir el nivell de risc en les primeres 24 hores d'ingrés. Les dades d'entrada seran variables relacionades amb simptomatologia i esdeveniments conflictius a l'últim mes. Per a tots dos casos, s'avaluarà el rendiment de diferents algorismes i es triarà el que millor caràcter predictiu tingui. Aquest model dual, que utilitza la Intel·ligència Artificial com a eina de suport a la decisió clínica, suposa un precedent en el control de riscos en models de portes obertes. Una vegada validat el model en la seva totalitat, seria escalable a altres unitats i permetria millorar la qualitat assistencial, mantenint la seguretat de totes les persones implicades en el procés de recuperació

En els darrers anys ha augmentat l'interès en la detecció automàtica d'emocions en branques com el neuromàrqueting, la psicologia i la sanitat, entre d'altres. Per això, s'han creat sistemes per identificar i analitzar l'estat emocional dels subjectes, i aquests es van desenvolupar a partir de diferents senyals fisiològics. En aquest estudi, es va fer servir senyal de seguiment ocular que conté informació oculomotora i de la dilatació pupil·lar obtinguda de manera no invasiva i econòmica. Amb el senyal enregistrat de 18 subjectes sans, es va estudiar la diferència de la resposta ocular entre l'existència o no d'estímuls audiovisuals, i entre la classe de vídeos que s'observen, de persones conegudes o desconegudes per als participants, així com les emocions que aquests generen. Per aconseguir-ho, es van entrenar diferents models d'aprenentatge automàtic i aprenentatge profund per dur a terme tres classificacions: entre presentar o no un estímul, entre que el vídeo sigui protag- onitzat per una persona coneguda pel subjecte o no, i, finalment, entre excitació alta i baixa de cada vídeo puntuat pel participant. Finalment, els resultats obtinguts en el primer cas van ser d'un 89.46% de precisió, en el segon cas, un 66.02%, i, en el tercer, un 67.59%, similars als que es van obtenir a la literatura. Amb això, es podríen buscar millors resultats a través d'una quantitat més gran de dades o mitjançant estudis multimodals amb altres tipus de senyals fisiològics. Addicionalment, es van testejar els millors models amb dades de pacient en síndrome de vigília sense resposta que van aconseguir 66.25% de precisió a la primera classificació i 62.95% a la segona. S'han obtingut resultats notables en models creats amb dades de subjectes sans. No obstant això, caldria en- trenar millors models amb dades de pacients, obtenint així millors resultats i poder treure conclusions sobre la resposta emocional dels pacients

La finalitat d'aquest projecte és desenvolupar un mètode per analitzar més profundament els patrons de marxa característics de pacients que presenten malaltia neurodegenerativa Polineuropatia Desmielinitzant Idiopàtica Crònica (CIDP), entorn aspectes morfològics i classificació de característiques de la marxa. Les dades s'han obtingut a través del Sistema Ephion Mobility i la realització de 2-Minute Walk Test (2MWT), a través de l'estudi clínic liderat per l'Hospital de Sant Pau de Barcelona. Les dades estudiades en el projecte consten de 30 tests de subjectes que participen en l'estudi clínic, a través del qual es desglossen en 60 tests per a l'avaluació dels valors assolits a través dels sensors de l'esquerra i dreta del cos del pacient. Les variables empleades s'obtenen concretament dels sensors, aplicats a les plantilles que s'incorporen en les sabates dels pacients per a l'execució del test funcional i els sensors aplicats a les articulacions dels turmells i els genolls. Es determina l'existència d'una diferència morfològica dels cicles de marxa predominant en pacients amb CIDP i s'avaluen patrons segons les característiques de marxa que aquests presenten. Alhora, es determinen els senyals i variables determinants per a la posterior classificació dels tests segons el nivell d'atàxia o steppage que els pacients presenten i es determina el mètode més robust per a la realització de la classificació de les dades

La qualitat de vida de les persones amb epilèpsia es veu considerablement alterada pel risc de patir crisis epilèptiques i els accidents que poden ocasionar durant el desenvolupament de les activitats quotidianes. És per això que s'han realitzat diversos intents per predir les crisis epilèptiques abans que es produeixin. Aquest projecte proposa fer ús del model Hierarchical Temporal Memory, una xarxa neuronal inspirada en el neocòrtex, per a la detecció d'anomalies al senyal EEG, concretament, l'activitat preictal prèvia a la crisi. Els principals avantatges d'aquesta innovadora xarxa neuronal són que permet l'aprenentatge en línia no supervisat de dades a temps real, així com l'adaptació als canvis de comportament de les dades d'entrada. Després de comprendre les bases del model i replicar-ne alguns exemples disponibles, s'ha creat un programa amb el model HTM, que torna com a valor de sortida la probabilitat d'anomalia, una mètrica que descriu la probabilitat que un registre sigui anòmal. L'objectiu és obtenir una resposta que ens permeti establir un llindar per diferenciar entre dades anòmales i no anòmales. S'han estudiat els paràmetres òptims del model i la seva influència a la sortida, sent la resolució i el període d'anomalia els que tenen un impacte més directe en el rendiment del model. A més, s'ha discutit com el remostreig és un pas crucial durant la fase de preprocessament per facilitar l'aprenentatge seqüencial de patrons de l'algorisme. Per a l'estudi, en primer lloc, s'ha avaluat el rendiment del model amb senyals EEG sintètics construïts a partir de dades reals de pacients, i s'ha pogut comprovar com el model és capaç de detectar la senyal durant la crisis epilèptica com a anormal, però no detecta amb èxit anomalies als registres de dades que precedeixen la crisi. També s'ha vist com el model aprenia les crisis com no anòmales després d'haver-ne vist una anteriorment, per la qual cosa s'han realitzat proves entrenant-se amb dades interictals, per posteriorment desactivar l'aprenentatge continu i passar dades que contenen crisi. Tot i els resultats satisfactoris, els valors d'anomalia instantània són molt elevats i fan dubtar si l'algorisme està aprenent els patrons interictals i si té una idea robusta del que és normal. Els assaigs realitzats amb dades de pacients reals han ratificat les conclusions anteriors. Per acabar, s'ha realitzat una comparació entre el model HTM i els Isolation Forests (IF). El model HTM no ha resultat avantatjós en comparació amb els IF, ja que aquest últim sembla detectar com a anòmals registres de dades abans de les crisis, però també suposa un risc més gran de falsos positius. Cap no dels dos ha demostrat ser un detector d'anomalies perfecte amb el conjunt de dades d'EEG utilitzat

To efficiently work with chromatography equipment, many laboratories end up choosing to implement solutions that include an automated sampling analysis system. From the user point of view, it is most important to make sure the instrument is always operative. When working with huge laboratories, with many instruments or where the offices are distant from said laboratories, checking the instruments can prove itself slow, bothersome, and insufficient. As a result of the curricular internship in the company Moyo Scientific, I was able to analyze this problem which led to the main idea of this project: to study the possible improvements in the managing of chromatography equipment to design a tool which allow to remotely check the instrument, through a device that can connect to the internet, like a mobile phone or a PC. With this project, I was able to thoroughly study the inner workings of Agilent's instruments. I've discovered the tools used by technicians and developers at Agilent to perform the maintenance and modifications of instruments, while I learned to use the tools myself to use them for my project.