KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 4 of 4

Some of the metrics are blocked by your
consent settings
PSD-EEGRepNet: A CNN Architecture with Multi-Branch RepBlocks for Power Spectral Density-Based Motor Imagery EEG Classification in BCI
(2025-01-01)
Thangthong, Kantham
;
Asadi, Fawad
;
Tungjitkusolmun, Supan
Motor Imagery (MI) based Brain-Computer Interfaces (BCIs) utilizing electroencephalography (EEG) offer significant potential, yet progress can be hindered by the computational demands of deep learning classifiers. This study introduces and evaluates a novel lightweight, multi-branch Convolutional Neural Network (CNN), inspired by efficient design principles, specifically for classifying MI tasks from Power Spectral Density (PSD) EEG features. Our objective was to achieve strong classification performance coupled with favorable development phase computational characteristics. Evaluated on 10 subjects from a public PhysioNet dataset using 5-fold cross-validation and two data overlap conditions (80%, 90%), the proposed model (~6.32M parameters, ~20 MFLOPs) demonstrated high mean classification accuracies (80.38% for 80% overlap, 85.69% for 90% overlap) and efficient training times (avg. 8.8s and 20.4s per fold, respectively). While performance scaled positively with data augmentation, inter-subject variability was noted. We conclude that the proposed architecture effectively balances high accuracy with significant offline computational efficiency offering a valuable tool for BCI research and a promising foundation for developing practical MI-BCI systems.
Some of the metrics are blocked by your
consent settings
Common Spatial Pattern Variants for Feature Extraction in Multi-Class Motor Imagery BCI
(2025-01-01)
Asadi, Fawad
;
Thangthong, Kantham
;
Tungjitkusolmun, Supan
Brain-Computer Interfaces (BCIs) based on Motor Imagery (MI) offer intuitive control but face challenges due to noisy electroencephalography (EEG) signals that often contain artifacts, necessitating effective feature extraction. Common Spatial Pattern (CSP) is a standard technique, yet its limitations in multi-class MI scenarios motivate various extensions. This paper investigates the classification performance, computational cost, and the crucial performance versus cost trade-off inherent in practical BCI applications of five prominent CSP variants, including standard CSP, Regularized CSP (RCSP), Filter Bank CSP (FBCSP), Sparse CSP, and Kernel CSP, when applied to multi-class MI classification. Using the PhysioNet MI dataset, we employed a consistent preprocessing pipeline (18 sensorimotor channels, filtering, epoching) and evaluated each variant by feeding extracted features into a Support Vector Machine (SVM) classifier with an RBF kernel. The evaluation revealed that standard CSP and RCSP achieved the highest peak average accuracy (~97.8%) with near-efficient computational cost. FBCSP yielded high accuracy (~96.7%) but incurred the highest computational cost. Sparse CSP was computationally efficient but achieved lower peak accuracy (~95.3%), while Kernel CSP showed the lowest peak accuracy (~92.8%) at moderate cost. The findings highlight the performance profiles and efficiency of these CSP methods for this multi-class MI task, indicating that standard CSP and RCSP offer a particularly effective combination of high accuracy and manageable computational load under the evaluated conditions.
Some of the metrics are blocked by your
consent settings
Evaluating synthetic neuroimaging data augmentation for automatic brain tumour segmentation with a deep fully-convolutional network
(2024-06-01)
Asadi, Fawad
;
Angsuwatanakul, Thanate
;
O'Reilly, Jamie A.
Gliomas observed in medical images require expert neuro-radiologist evaluation for treatment planning and monitoring, motivating development of intelligent systems capable of automating aspects of tumour evaluation. Deep learning models for automatic image segmentation rely on the amount and quality of training data. In this study we developed a neuroimaging synthesis technique to augment data for training fully-convolutional networks (U-nets) to perform automatic glioma segmentation. We used StyleGAN2-ada to simultaneously generate fluid-attenuated inversion recovery (FLAIR) magnetic resonance images and corresponding glioma segmentation masks. Synthetic data were successively added to real training data (n = 2751) in fourteen rounds of 1000 and used to train U-nets that were evaluated on held-out validation (n = 590) and test sets (n = 588). U-nets were trained with and without geometric augmentation (translation, zoom and shear), and Dice coefficients were computed to evaluate segmentation performance. We also monitored the number of training iterations before stopping, total training time, and time per iteration to evaluate computational costs associated with training each U-net. Synthetic data augmentation yielded marginal improvements in Dice coefficients (validation set +0.0409, test set +0.0355), whereas geometric augmentation improved generalization (standard deviation between training, validation and test set performances of 0.01 with, and 0.04 without geometric augmentation). Based on the modest performance gains for automatic glioma segmentation we find it hard to justify the computational expense of developing a synthetic image generation pipeline. Future work may seek to optimize the efficiency of synthetic data generation for augmentation of neuroimaging data.
Some of the metrics are blocked by your
consent settings
Neural correlates of face perception modeled with a convolutional recurrent neural network
(2023-04-01)
O’Reilly, Jamie A.
;
Wehrman, Jordan
;
Carey, Aaron
;
Bedwin, Jennifer
;
Hourn, Thomas
Objective. Event-related potential (ERP) sensitivity to faces is predominantly characterized by an N170 peak that has greater amplitude and shorter latency when elicited by human faces than images of other objects. We aimed to develop a computational model of visual ERP generation to study this phenomenon which consisted of a three-dimensional convolutional neural network (CNN) connected to a recurrent neural network (RNN). Approach. The CNN provided image representation learning, complimenting sequence learning of the RNN for modeling visually-evoked potentials. We used open-access data from ERP Compendium of Open Resources and Experiments (40 subjects) to develop the model, generated synthetic images for simulating experiments with a generative adversarial network, then collected additional data (16 subjects) to validate predictions of these simulations. For modeling, visual stimuli presented during ERP experiments were represented as sequences of images (time x pixels). These were provided as inputs to the model. By filtering and pooling over spatial dimensions, the CNN transformed these inputs into sequences of vectors that were passed to the RNN. The ERP waveforms evoked by visual stimuli were provided to the RNN as labels for supervised learning. The whole model was trained end-to-end using data from the open-access dataset to reproduce ERP waveforms evoked by visual events. Main results. Cross-validation model outputs strongly correlated with open-access (r = 0.98) and validation study data (r = 0.78). Open-access and validation study data correlated similarly (r = 0.81). Some aspects of model behavior were consistent with neural recordings while others were not, suggesting promising albeit limited capacity for modeling the neurophysiology of face-sensitive ERP generation. Significance. The approach developed in this work is potentially of significant value for visual neuroscience research, where it may be adapted for multiple contexts to study computational relationships between visual stimuli and evoked neural activity.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results