[dipl] Perustieteiden korkeakoulu / SCI

Permanent URI for this collectionhttps://aaltodoc.aalto.fi/handle/123456789/21

Browse

Browsing [dipl] Perustieteiden korkeakoulu / SCI by Degree programme/Major subject "Biomedical Engineering"

Filter results by typing the first few letters
Now showing 1 - 20 of 81
  • 3D geometric correction in 2D multi-slice acquired MR images used in radiation therapy planning
    (2020-03-16) Haukipuro, Eeva-Sofia
     Perustieteiden korkeakoulu | Master's thesis
    The role of MRI in the external radiation therapy treatment has increased in the past few decades significantly due to its superior soft-tissue contrast and absence of ionizing radiation. The two challenges concerning the lack of electron density values and limited geometric accuracy to use MR images in RT have been overcome. MR images can either be aligned with CT images or converted to pseudo-CT images for radiation dose planning, and the required geometric accuracy can be achieved by optimizing sequences and utilizing geometric correction algorithms. The purpose of this thesis was to improve the geometric accuracy of MR images by implementing 3D geometric correction to 2D multi-slice acquired images. Until now, this feature has not been available in Philips MR scanners. The advantages of 2DMS acquired images over 3D images are the reduced sensitivity to motion artifacts and the contrast that is preferred by radiologists. The geometric correction algorithm itself was not modified, since the existing algorithm for 3D images could be reused. The performance and the geometric accuracy achievable with the algorithm were evaluated with phantom imaging and the effect of the algorithm on human anatomy was evaluated with volunteer imaging. Altogether 21 subjects were imaged with Ingenia 1.5 T, Ingenia 3 T and 1.5 T MR scanner of Elekta Unity. The results reveal that the geometric distortion is significantly reduced when through-plane correction is applied alongside in-plane correction. The greatest effects are shown on the edges of the field-of-view where the effect of gradient non-linearities are the largest. It was also proven that the algorithm performs the geometric correction almost as well in 2DMS as in 3D images.
  • Adapting a Multi-Stage and Multi-Label Vertebra Segmentation Pipeline for Segmenting Teeth from CBCT images
    (2024-08-19) Laine, Veera
     Perustieteiden korkeakoulu | Master's thesis
    Automatic teeth segmentation plays a crucial role in digital dentistry as it enables automatization of workflows as well as improves treatment planning and communication with patients. However, achieving a fully automatic segmentation of teeth is difficult due to e.g. high presence of restorative material-based artefacts, anatomical variability, and similar densities and proximity of maxilla, mandible and teeth. This thesis proposes an automatic 33-class semantic teeth segmentation pipeline by adapting an open-source three-step vertebra segmentation pipeline from MONAI Label. The adapted vertebra pipeline was based on the convolutional neural network SegResNet. A total of 30 CBCT scans were selected and annotated for training (20 scans), validation (2 scans), and testing (8 scans). The dataset was heterogeneous containing implants, missing teeth, 3rd molars, and artefacts. Total processing times, segmentation accuracy (IoU), teeth classification accuracy, and memory consumption were used for evaluating the model performance. The final iteration of the pipeline achieved a teeth classification accuracy of 98.0 % and an average IoU of 84.6 % (SD 14.2 %). The total processing times were on average 41.52 s on the CPU and 4.32 s on the GPU for FOV 80.2 mm2 volume size which enables the pipeline to be used in a suitable time frame even without high hardware specifications.
  • Anatomical Segmentation of CT images for Radiation Therapy planning using Deep Learning
    (2018-10-08) Schreier, Jan
     Perustieteiden korkeakoulu | Master's thesis
    Radiation therapy is one of the key cancer treatment options. To avoid adverse effect in tissue surrounding the tumor, the treatment plan needs to be based on accurate anatomical models of the patient. In this thesis, an automatic segmentation solution is constructed for the female breast, the female pelvis and the male pelvis using deep learning. The deep neural networks applied performed as well as the current state of the art networks while improving inference speed by a factor of 15 to 45. The speed increase was gained through processing the whole 3D image at once. The segmentations done by clinicians usually take several hours, whereas the automatic segmentation can be done in less than a second. Therefore, the automatic segmentation provides options for adaptive treatment planning.
  • Application of Computational Histopathology to Quantify Immune Cells in Clear-cell Renal Cell Carcinoma Tissue
    (2022-01-17) Brummer, Otso
     Perustieteiden korkeakoulu | Master's thesis
    Spatial arrangement and number of lymphocytes in biopsy textures have been shown to influence the prognosis and the treatment efficiency in cancers. In particular, the intratumoral lymphocyte infiltration generally correlates positively with prognosis. In the case of renal cell clear cell carcinoma however, the association is ambiguous. Previously infeasible clinical problems, such as the lymphocyte infiltration quantification, may be computationally feasible with the application of modern convolutional neural networks. This thesis aims to produce a combination of two quantification networks which can recognize the texture and number of lymphocytes in whole slide renal cell clear cell carcinoma nephrectomy samples. Moreover, a preliminary attempt at induction of the immune response phenotypes from the results is made. An annotated dataset consisting of 36 000 labelled texture images and a regional dataset of 200 images with labelled lymphocytes and general cells were produced from the publicly available renal cell carcinoma hematoxylin-eosin-stained biopsy scans of Cancer Genome Atlas image archive. These datasets were utilized to train two convolutional networks. One to evaluate the tissue texture and one to evaluate the type and counts of cells in an image. The tissue texture evaluation network achieved total accuracy of 90 % in a collected testing dataset. Often the tissue is a combination of multiple textures and defining a single label for it without a degree of ambiguity is impossible. Taking this into account, the results are more than satisfactory. The cell classification network achieved VOC AUC 0.5 of 62.3 % for non-lymphocytes and 48.7% for lymphocytes. Cell classification suffers considerable from the variance between different images. Localization of the cell matches is satisfactory, but the differentiation of lymphocytes from other cells mainly depends on the size and darkness of the cell in a stain which has an amount of volatility across different staining protocols. Otherwise, the results are comparable to previous studies conducted on the immune cell counts. This thesis indicates that computational analysis of histopathological images is possible even with consumer grade systems. A set of stain color normalization solutions could be developed to reliably produce uniform images to ease the challenging lymphocyte classification problem.
  • Application of linear machine learning methods for the diagnosis of mild traumatic brain injuries
    (2022-05-16) Aaltonen, Juho
     Perustieteiden korkeakoulu | Master's thesis
    Diagnosis of mild traumatic brain injury (mTBI) is challenging regardless of the high number of cases worldwide. Structural imaging findings are often lacking, and the aberrations in behavior are not specific to mTBI. Most mTBI patients recover rapidly without any prolonged symptoms, but 10-15% suffer from prolonged symptoms. Neurophysiological studies have demonstrated abnormal slow-wave (< 7 Hz) activity in mTBI compared with healthy controls when measured early after the trauma, but the analysis requires specific expertise. In this master's thesis, linear machine learning methods' ability to separate mTBI patients from healthy controls based on their magnetoencephalographic (MEG) brain activity is studied. Three widely used machine learning methods were used: linear discriminant analysis (LDA), support vector machine (SVM), and logistic regression (LR). The machine learning methods were applied on resting-state MEG power spectra (1-88 Hz) from two independent datasets. The results showed that it is possible to separate mTBI patients and healthy controls from each other based on their power spectra with median accuracy of 80-90%. There was no significant difference between the used machine learning methods, and the results were consistent between the two datasets. This suggests good performance of easily applicable linear machine learning methods also for clinical use for finding the patients who may benefit from close follow-up during the recovery period.
  • Artifact detection in neonatal EEG using unsupervised machine learning
    (2017-04-03) Kauppila, Minna
     Perustieteiden korkeakoulu | Master's thesis
    The neonatal electroencephalogram (EEG) is an important tool for assessing cortical activity in critically ill neonates. However, the EEG is often contaminated by artifacts such movement artifacts, electromyographic (EMG) artifacts from muscle activity and electrocardiographic (ECG) artifacts. These artifacts make visual inspection difficult and negatively influence the results of automated analysis. Even though several methods for automated neonatal EEG analysis have been developed, there is a significant lack of comprehensive artifact detection systems for the neonatal EEG. In this thesis, an automated artifact detection system based on a semi-supervised Gaussian mixture model (GMM) is presented. We examined the effects of feature set size, mixture number and the use of principal component analysis (PCA) as a pre-processor. Performance was assessed using the area under the receiver operating characteristic (AUC) and estimated using leave-one-patient-out cross-validation. The best-performing system was obtained with 23 features, 30 mixtures and no PCA (median AUC = 0.95, IQR: 0.83--0.99). EMG and movement artifacts were detected with the highest accuracy.
  • Auditory evoked responses of domestic cats measured using magnetoencephalography
    (2024-11-18) Henttonen, Markus
     School of Science | Master's thesis
    Magnetoencephalography (MEG) is a non-invasive method to measure the electrical activity of the brain based on the magnetic fields produced by neurons. MEG is applied in humans for basic neuroscience and clinical diagnostics. In contrast, invasive electric measurements have been performed in model animals; for example, domestic cats (felis catus) have been used for a long time as model organisms in neuroscience. Here, we present a non-invasive MEG measurement in cats by measuring their auditory evoked responses. These measurements can help bridge the gap between the invasive measurements in cats and the non-invasive measurements in humans. MEG has traditionally been performed using superconducting SQUID sensors arranged in a helmet designed for humans, but these systems are impractical for smaller animals such as cats. Hence, we applied novel optically pumped magnetometers (OPMs) that can be placed in a custom array suitable for cats. We measured 8 pet cats (6 females, 2 males) with a mean age of 5.4 years. The stimulus was based on the so-called local–global paradigm comprising a standard (75%) and deviant (25%) series of five 50-ms tone pips at 1 kHz and 50 kHz. The measurement lasted maximum of 40 min not including breaks. Data from 6 out of 8 subjects were used for the analysis. The equivalent of the human N100 auditory response was detected at around 65 ms after the onset of the tone pips. A likely human mismatch-negativity analogue was seen at 65 ms as a stronger N100 after the ultrasound tone. However, convincing evidence for the P300 response was not found. These measurements show that MEG-based non-invasive neuroscience is feasible in animals. The measured latencies of N100 and probable mismatch negativity are in line with previous studies, which indicates the measurements have been successful.
  • Automated Solutions in Chemical Quality Control
    (2024-01-22) Salojärvi, Pirta
     Perustieteiden korkeakoulu | Master's thesis
    Laboratory automation is growing in demand as laboratories seek benefits such as enhanced efficiency and reduced human errors. Trained personnel can be allocated for more meaningful and demanding complex tasks hence bringing cost benefits. Automation enables real-time sample tracking ensuring more up-to-date information. Digitized records of the analysis decrease the risks of misinformation and lost information providing higher quality throughout the entire process. Despite the advantages, laboratory automation implementation has challenges such as high costs, intricate method transfers, and lengthy validation processes. The challenges described are mainly caused by the technology being relatively new for companies and hence there isn’t a lot of experience in implementation. Chemical quality control laboratories in regulated context, have to follow the regulations that control the safety of the produced products. The regulations, such as Current Good Manufacturing Practices, control the whole production chain, which the quality control laboratories are a part of. The automated processes have to follow these same regulations. Finnish pharmaceutical company, Orion, is launching a project for updating the sample and eluent preparation methods for tablet samples that are analyzed with High-Performance Liquid Chromatography in the chemical quality control laboratory. The current preparation method is manual and the target is to automatize it. An increase in efficiency is sought from automated solutions. This thesis presents Orion’s situation, detailing the manual processes and the challenges posed by the absence of automation. Orion explores potential automation solutions for sample preparation, including Sotax TPW, Beckman Coulter Biomek i7, Chemspeed Swing SP, and Accroma SamplePrep System. Potential options for automating the eluent/buffer preparation are Chemspeed Flex EP and LabMinds REVO. The options are evaluated based on their compatibility with Orion’s requirements and their profitability based on the investment compared to the efficiency improvement provided by the device. Each of these possibilities would bring benefits to the current manual process. However, if Orion decides to apply automation in the laboratory, they would face the same challenges as all the other companies implementing laboratory automation for a highly regulated quality control environment. Precise planning early on the investment project is the key to tackling the challenges. This thesis contributes insights into the challenges and opportunities associated with laboratory automation in a highly regulated quality control environment.
  • Clinical evaluation of a proof-of-concept transcranial magnetic stimulation system under the Medical Device Regulation
    (2020-08-17) Lowe, Anastasia
     Perustieteiden korkeakoulu | Master's thesis
  • Co-Culture System Development for Endothelial Cells and Cardiomyocytes
    (2019-09-30) Antola, Laura
     Perustieteiden korkeakoulu | Master's thesis
    The human heart consists of several different cell types, the main types being cardiomyocytes and endothelial cells. Different cell types interact constantly with each other, which affects their function. This interaction and its effects were studied by developing a co-culture system, in which induced pluripotent stem cell derived cardiomyocytes and endothelial cells were cultured together as a co-culture. Additionally, in the inner lining of blood vessels endothelial cells are constantly exposed to shear stress caused by blood flow. This kind of environment was simulated by culturing endothelial cells on a slide that was attached to a flow system, in which a pump pushed medium to move through the slide. The co-cultures and flow-cultures were studied using fluorescence microscopy and single-cell RNA-sequencing. Cardiomyocytes and endothelial cells were viable in co-cultures, and flow made endothelial cells to align parallel to the flow direction. Culturing iPS-derived endothelial cells together with iPS-derived cardiomyocytes increased expression of arterial and venous genes in iPS-endothelial cells but had little effect on gene expression of iPS-cardiomyocytes. Shear stress caused by flow evoked a healthier endothelial cell phenotype, since it upregulated genes that affect important endothelial functions such as vascular tone regulation and angiogenesis.
  • A Compact Device for Ultrasound-Enhanced Biopsy
    (2019-06-18) Mikkola, Matti
     Perustieteiden korkeakoulu | Master's thesis
    Fine-needle aspiration (FNA) biopsy is a safe, non-invasive method used in sampling tissue and fluids for diagnosis. However, up to a third of FNA biopsies fail due to insufficient sample volume. This can be remedied by coupling ultrasound to the needle, which increases the sample volume up to five times more and could also make the procedure less painful due to reducing insertion resistance. An ultrasound-enhanced fine needle biopsy (USeFNB) device was recently built for this purpose. The device is adequate for research, but unnecessarily bulky and complex for use in a clinical environment. In this thesis, a novel ultrasound device was designed and built with the goal of reaching similar performance to the USeFNB system, but with the intention to miniaturize the instrument size. The theory of solid mechanics as well as finite element analysis were used to determine dimensions of the system elements. A prototype was built to these specifications. The needle displacement and cavitation activity produced by the prototype were characterized using high speed camera footage and compared to the USeFNB system. With the same electrical input power, the prototype showed somewhat inferior performance in terms of these quantities. The biopsy yield was not studied due to time constraints. Considering these results, there is reason to believe that enhanced yield could be obtained as well, though some modifications might be required.
  • Compatibility and safety testing of submodules for a novel fMRI-compatible mTMS device
    (2020-08-17) Kanerva, Niina
     Perustieteiden korkeakoulu | Master's thesis
    Multi-locus transcranial magnetic stimulation (mTMS) and functional magnetic resonance imaging (fMRI) provide complementary approaches to the study of brain. Combining these methods into an mTMS–fMRI hybrid allows conducting studies where the brain is simultaneously stimulated and monitored. The constituent parts of the novel fMRI-compatible mTMS system, called submodules, need to be tested for safety and compatibility. The aim of this thesis was to ensure the safety and compatibility for the selected submodules of the mTMS–fMRI hybrid. To meet this aim, this master’s thesis composes test protocols and presents guidelines for four different submodules of the system. The protocol tests were created to simulate both normal and fault condition scenarios, and the guidelines chosen to verify compliance with limits set for normal operation. The protocol instructions are composed to allow even an inexperienced user to conduct the tests and assess safety and compatibility of the tested device. The protocol tests were able to identify needs for submodule revisions, need for a change in manufacturing process of one submodule, incorrectly installed components, and a faulty component. It was assessed that some of the identified faults could had let to safety and compatibility issues. After the submodules were revised based on protocol test results, they eventually passed the test protocols. This indicates that the test protocols can serve as tool to identify faulty behaviour and can be utilized to assess the safety and compatibility of each submodule.
  • Deep Learning Methods for Visual Fault Diagnostics of Dental X-ray Systems
    (2018-08-20) Agrawal, Harshit
     Perustieteiden korkeakoulu | Master's thesis
    Dental X-ray systems go through rigorous quality assurance protocols following their production and assembly. The protocols include tests, which address the image quality and find certain errors or artifacts that may be present in the images. Detecting faults from the images require human effort, experience, and time. Recent advances in deep learning have proven them to be successful in image classification, object detection, machine translation. The applications of deep learning can be extended to fault detection in X-ray systems. This thesis work consists of surveying, applying, and developing state-of-art deep learning approaches for detection of visual faults or artifacts in the dental X-ray systems. In this thesis, we have shown that deep learning methods can detect geometry and collimator artifacts from X-ray images efficiently and rapidly. This thesis is a precursor for further development of deep learning methods to include detection of wide range of faults and artifacts in X-ray systems to ease quality assurance, calibration, and device maintenance.
  • Deep learning text-to-speech synthesis with Flowtron and WaveGlow
    (2023-05-15) Sairanen, Veera
     Perustieteiden korkeakoulu | Master's thesis
    Innovation in the field of artificial speech synthesis using deep learning has been rapidly increasing over the past years. Current interest lies in the synthesis of speech that is able to model the complex prosody and stylistic features of natural spoken language using a minimal amount of data. Not only are such models remarkable from a technological perspective they also have immense potential as an application of custom voice assistive technology (AT) for people living with speech impairments. However, more research should be focused on the evaluation of the applicability of deep learning text-to-speech (TTS) systems in a real-world context. This thesis aims to further this research by employing two well-known TTS frameworks, Flowtron and WaveGlow, to train a voice clone model on limited personal speech data of a person living with locked in syndrome (LIS). The resulting artificial voice is assessed based on human perception. In addition, the results of the model are showcased in a user-friendly TTS application that also acts as a prototype for custom voice AT. Through the work in this thesis we explore the fascinating world of deep learning based artificial speech synthesis and inspire further research in its relevance toward the development of inclusive technology.
  • Designing a pilot experiment to study dissimilarity of neural activation detected by MEG and fMRI when imaging language processes
    (2024-09-04) Nielikäinen, Jonna
     School of Science | Master's thesis
    Multiple functional brain imaging methods, such as electroencephalography (EEG), magnetoencephalography (MEG), and functional magnetic resonance imaging (fMRI), are commonly used to study the activity of the human brain. These methods rely on different physiological phenomena to capture information about neural activity, and their findings may not always align, even when they are used to image the same brain processes. Differences have been found when comparing MEG and fMRI measurements of the same high-level cognitive tasks, such as reading. The reason for these discrepancies is not fully understood. However, differences might have to do with the distinct nature of the electrophysiological signals measured by MEG and the hemodynamic signals measured by fMRI. The relationship between hemodynamic signals and neural activity is complex. Therefore, the two modalities could be differently sensitive to some aspects of neural activation, such as varying top-down and bottom-up processes. Understanding better what causes the difference in the activation seen by MEG and fMRI would allow better spatiotemporal insight into how the brain works. Carefully designed experimental paradigms are needed to test the relationship between the signals measured by these imaging modalities. This thesis aimed to design an experimental paradigm for an imaging experiment to study the relationship between MEG and fMRI during language processing tasks. The main aim of the thesis was to determine a suitable parameter that could be tuned to increase/decrease the similarity between the MEG and fMRI imaging results and create an experimental setup where the effects of this parameter could be systematically investigated. A comprehensive literature review was performed on which to base the decisions regarding the experimental paradigm. Sentences with varying syntactic and semantic structures were chosen to alter the similarity between MEG and fMRI data. Furthermore, a pilot measurement using MEG was performed. The results seem promising. They show differences in responses when the semantic and syntactic structures of the sentences are modified, indicating that the experimental design works. The next step will be to continue with an fMRI pilot measurement, which will allow a comparison of the results of MEG and fMRI to see whether they reveal the postulated graded dissimilarities between the two imaging modalities.
  • Designing and building mTMS transducers for bihemispheric brain stimulation
    (2024-01-23) Nurmi, Samuel
     Perustieteiden korkeakoulu | Master's thesis
    Transcranial magnetic stimulation (TMS) is a non-invasive method for stimulating the brain. In TMS, a strong current pulse is applied through coils to generate a magnetic field that induces a stimulating electric field in the brain. TMS has many applications in neuroscience and therapy, including functional brain imaging. TMS is applied by placing the stimulation coil manually over the patient’s scalp on the desired stimulation region. Thus, the translation and rotation of the stimulating electric field are achieved by repositioning the coil manually. Recently, multi-locus TMS (mTMS) has been developed, wherein instead of a single stimulation coil, multiple coils are used simultaneously. This allows the user to electronically alter the stimulation location and orientation, providing an advantage over conventional TMS. The stimulation can be altered on a millisecond timescale, enabling the study of connectivity between different brain regions. Existing mTMS transducers are flat, which decreases the coil’s coupling to the brain due to the windings being further away from the brain on the edges. A curved three-dimensional transducer would bring the coil windings closer to the brain and thus improve the coupling. This thesis aims to design a curved mTMS transducer, specifically tailored for paired motor cortex to motor cortex stimulation, where two transducers are placed on the head, one on each hemisphere. We constructed two different prototypes of this transducer. The designed mTMS transducer consist of three coils: two figure-of-eight coils and one round coil. The two figure-of-eight coils had a good correspondence to the computational model with respect to the stimulation strength. The third round coil had a smaller stimulation strength than computationally modelled. The field patterns and field directions corresponded between the prototype and computational model. A second prototype of the mTMS transducer was built to enhance electrical safety and allow for applications in studies with healthy participants. In addition to the new transducer prototypes, the other outcome of this thesis for the coil design process was a problem-specific approach through three-dimensional curved coil geometry. Where the coil was tailor-made for a specific purpose for maximal efficiency and functionality of the transducer.
  • Detecting fatigue from nocturnal heart rate variability
    (2018-01-22) Santaniemi, Nuutti
     Perustieteiden korkeakoulu | Master's thesis
    The focus of this thesis was to study if the level of fatigue could be assessed via nocturnal heart rate variability (HRV). HRV is a phenomenon that can be used to evaluate the status of the autonomic nervous system. A commonly used method for assessing the level of fatigue via HRV is an orthostatic test. The test is widely used by athletes in order to maintain the right balance between training and recovery. The test should be done every day as soon as possible after waking up, and therefore it may seem too laborious for nonprofessional athletes. Evaluating fatigue via nocturnal HRV would make monitoring fatigue less laborious by automating the measurement. The data that was used in this study consisted of 234 overnight recordings of RRi (time interval between adjacent heart beats) and 234 orthostatic test from 11 different persons. The results of the orthostatic test were then compared to different HRV variables calculated from the nocturnal RRi measurements. The nocturnal HRV variables were calculated from two different segments from the RRi data. The results reveal that the standard deviation of the RRi (std RRi) from the estimated slow-wave sleep (SWS) segment correlated with the mean RRi in the orthostatic test in standing position. The average correlation was minor (τ = 0.26 ± 0.26) and it was calculated with non-parametric Kendall’s Tau test. Out of 11 participants, the correlation was statistically significant (p < 0.05) for eight participants. It was concluded that the std RRi can not yet replace the orthostatic test in fatigue analysis. Despite the significant correlation, further analysis on the similarity of the std RRi and the orthostatic test should be conducted. Also due to interindividual differences, the suitability of this nocturnal parameter needs to be verified for each individual before one can consider leaving out the orthostatic test.
  • Development and comparison of EEG analysis pipelines for diagnosis of mild traumatic brain injuries
    (2023-05-15) Porta, Estanislao
     Perustieteiden korkeakoulu | Master's thesis
    Mild Traumatic Brain Injuries (mTBI) are very common and can have a significant impact on an individual’s life. The symptoms of mTBI, which include headache, fatigue, sleep alterations, dizziness and difficulty in concentrating, may be more severe than the injury’s "mild" label suggests. Although most patients recover promptly, one in eight will suffer prolonged symptoms for weeks, months or possibly, throughout their whole lifetime. Understanding the mechanisms of mTBI is important for promoting individual health and well-being, but diagnosing remains challenging due to the large individual variability in symptoms, and the limitations of brain imaging techniques. Identifying objective biomarkers could increase diagnostic accuracy and improve clinical outcomes. Studies indicate that computational methods combined with neuroimaging could increase the diagnostic accuracy, by recognising subtle abnormalities in neurophysiological activity thus differentiating mTBI patients from healthy subjects. Machine learning techniques are frequently used in computational pipelines, and as these pipelines become more complex, they start to resemble software packages. To ensure that the results are reliable and the workflows are reproducible, it is important to follow computational best practices. In this Master’s Thesis, a well-documented software package for classifying mTBI patients was designed and developed, following best standards from computational research. A pipeline was constructed to process, analyse, and classify electroencephalographic (EEG) data applying a reproducible approach, based on different machine learning models. The software has been made publicly available at GitHub as a Python package: https://github.com/BioMag/mtbi_meeg. The Thesis describes the software development process of the pipeline, along with the performance metrics from different classifiers when utilising EEG data of 35 mTBI patients and 36 healthy controls, measured in BioMag Laboratory, Helsinki University Hospital. The pipeline can be further expanded to incorporate ther imaging modalities for more comprehensive analyses.
  • Development and validation of source-estimation methods for stereoelectroencephalography
    (2024-05-20) Simanainen, Santeri
     Perustieteiden korkeakoulu | Master's thesis
    Stereoelectroencephalography (SEEG) is an invasive clinical procedure in which brain-activity-related fluctuations in the intracranial electric potential are measured with depth electrodes. This method allows higher signal-to-noise ratio and spatial resolution compared to non-invasive alternatives and is used in preoperative evaluation for surgical treatment of medically refractory epilepsy. Traditionally, the clinical evaluation is based on a sensor-level assessment of the seizure onset. However, as the spatial sampling of SEEG is sparse, this information can sometimes be insufficient for planning surgical resection. By encompassing biophysical laws about signal generation, source estimation could expand the narrow field of view of SEEG, thus improving localization accuracy and reducing the likelihood of inconclusive findings. This work aimed to develop a source-estimation pipeline for SEEG and validate the approach by modeling intracranially recorded somatosensory evoked potentials (SSEPs) following median nerve stimulation. The data used in the work were provided by the Helsinki University Hospital, including the SEEG recordings, structural magnetic resonance images, and post-operative computational tomographic images from 15 patients. The forward problem associated with SEEG was solved by creating individualized volume conduction models with realistic skull and electrode geometries and computing the field potential numerically using the boundary element method. To validate the approach, the early SSEP responses were modeled as equivalent current dipoles (ECDs). The localization error of the ECD estimates for sources in the implanted hemispheres ranged between 2.5 and 15 mm, with a median error of 7.1 mm. On the hemisphere without electrode coverage, no accurate localization was possible, the median error being 62 mm. Overall, the error was dependent on the source-implantation distance and the signal-to-noise ratio, but at distances less than 3 cm the error was likely dominated by forward modeling errors. The study concluded that even a relatively simple model can accurately localize neuronal sources that are in the vicinity of the depth electrodes but could not be delineated by visual inspection of the sensor-level signals.