Smadzeņu-datora interfeiss rehabilitācijai pēc insulta, izmantojot EEG signālus un virtuālās realitātes brilles

Abstract

Šis bakalaura darbs pēta uz EEG balstītu smadzeņu-datora interfeisa potenciālu insulta rehabilitācijā, izmantojot funkcionālo neiroplastiskumu. Pētījumā tiek pieminēta rehabilitācijas efektivitātes uzlabošanu, pievienojot virtuālās realitātes (VR) brilles, ar mērķi nodrošināt insulta pacientiem īstajai dzīvei pietuvinātāku pieredzi. Galvenais uzsvars tiek likts uz optimālas datu priekšapstrādes, normalizēšanas un klasifikācijas metožu kombinācijas noteikšanu, lai sasniegtu visprecīzāko mašīnmācīšanās modeli EEG signālu apstrādei. Darbā tiek izmantota OpenBCI EEG ierīce. Pētījumā tika izvērtēta izstrādātā modeļa veiktspēja, klasificējot EEG signālus, kas satur labās rokas kustības, izmantojot vairāku failu un reālā laika datus. Lai gan precizitāte atsevišķos ierakstos atšķiras, vidējā klasifikācijas precizitāte sasniedza 88,3%, kad ārējie faktori bija līdzīgi (diapazons: 72,15% - 95,86%). Tomēr ārējie faktori, piemēram, ierakstīšanas apstākļi, elektromagnētiskais lauks un savienojums starp galvu un elektrodiem, dažreiz rada nesalīdzināmus datus starp dažādām eksperimenta reizēm, kad šie faktori bijuši atšķirīgi. Tas paliek galvenais izaicinājums un joma, kas prasa turpmāku izpēti un uzlabojumus. Šie rezultāti norāda uz EEG balstīta smadzeņu-datora intefeisa un VR tehnoloģijas potenciālu uzlabot insulta rehabilitācijas rezultātus un pavērt ceļu personalizētākai un efektīvākai rehabilitācijai. Keywords: Smadzeņu-datora interfeiss, EEG signāli, insulta rehabilitācija, virtuālās realitātes (VR) brilles

Stroke is a prevalent and life-altering medical condition, affecting millions of people worldwide each year. In Latvia alone, thousands of individuals experience a stroke annually, making it a significant health concern. Motor dysfunction, a common consequence of stroke, can lead to permanent disability and significantly impact the quality of life for survivors. Despite undergoing intensive physical rehabilitation, a significant proportion of stroke patients continue to experience limitations in motor function. The brain's ability to change, plasticity, offers hope for individuals seeking to regain control over paralyzed limbs, with various rehabilitation approaches focusing on mental imagery, tactile stimulation, passive movements, and mirror therapy. However, these methods are primarily suitable for patients with some residual limb function, leaving fewer options for those with full paralysis. In addition, patients can get unmotivated to continue rehabilitation as the same movement has to be repeated many times. Recent advancements in neuroscience and technology have opened the way for innovative rehabilitation approaches, particularly using electroencephalography (EEG)-based Brain-Computer Interfaces (BCIs). By connecting individuals to EEG device and decoding their brainwaves through machine learning algorithms. This thesis uses OpenBCI EEG device to investigate the potential of EEG-based BCIs in stroke rehabilitation. The study mentions improvement of rehabilitation effectiveness by incorporating virtual reality (VR) glasses, aiming to provide stroke patients with a more realistic experience. A key focus is identifying the optimal combination of data preprocessing, normalization, and classification techniques to achieve the most accurate machine learning model for EEG signal processing. The research evaluated the performance of the developed model in classifying EEG signals associated with right-hand movements using data from multiple file recordings and real-time data. While accuracy varied across individual recordings, the average classification accuracy reached 88.3% when external factors were similar (range: 72.15% - 95.86%). This was achieved by a combination of a notch filter of 50Hz, a bandpass filter from 5 to 50Hz, z-score normalization and LSTM deep learning algorithm. However, external factors such as variations in recording conditions, electromagnetic fields, and connections between the head and electrodes introduce inconsistencies in the recorded data across different sessions. While z-score normalization achieved promising results when these factors were similar, further refinement is necessary. Normalization could be the step where data could be adjusted to become comparable session-to-session, which is crucial for improvement. Various techniques, including common average referencing, adaptive filtering, and calibration periods, were explored to address this issue, but none achieved improvements in accuracy across all types of data influenced by external factors. This remains the primary challenge and area requiring further investigation and improvement. These findings underscore the potential of EEG-based BCIs coupled with VR technology to enhance stroke rehabilitation outcomes and pave the way for more personalized and effective rehabilitation. Keywords: Brain-Computer Interface (BCI), EEG signals, stroke rehabilitation, virtual