Ömer Muhammet Soysal

Electroencephalography (EEG) signals provide a unique opportunity for personal identification due to their inherent permanence and uniqueness. However, EEG data are complex, multidimensional, non-stationary, and time-dependent, making feature extraction and classification particularly challenging. This study proposes a framework by integrating autoencoder-based feature extraction with self-attention mechanism for EEG-based personal identification utilizing the proposed EEG data cube spatio-temporal stream representation. Three types of stimuli were applied: auditory, cognitive, and resting state. EEG data were collected in-house from college students in two sessions separated by 10 days. A longitudinal design was employed, with Session-1 data used for training and Session-2 data for testing, ensuring robustness against inter-session variability. Extracted features were subsequently classified using Support Vector Machines (SVM), Artificial Neural Networks (ANN), k-Nearest Neighbors (KNN), and Random Forests (RF). SVM achieved accuracy scores 97.62% when identifying two subject, 82.54% when identifying five subjects and 62.90% when identifying seven subjects using the resting state neural patterns. This research highlights the feasibility of using deep learning networks strengthened with a self-attention mechanism for advancing biometric identification through EEG signals.

Individual brain activity patterns derived from electroencephalogram (EEG) data offer a unique source for personal identification, introducing a novel approach to the field. Autoencoders are well-known machine learning models that automate feature extraction, which is a crucial step in biometric identification. Among various types of autoencoders, the domain-adaptive autoencoder (DAAE) is explored for feature extraction. The extracted latent features are employed by four machine learning classifiers, KNN, ANN, SVM and RF, for personal identification. Two domain adaptation approaches were presented. The proposed frameworks were evaluated in a longitudinal setting, using three types of EEG recordings: resting state, auditory and cognitive stimuli. Model performance was assessed through experiments involving seven-, five- and two-subject classification tasks. The highest identification accuracy, 100%, was achieved by the SVM-based model in the two-subject experiment, using features extracted with the uniform referential DAAE. Similarly, the RF-based model attained an accuracy of 99.84% in the two-subject experiment when trained on features obtained from the softmin referential DAAE. As expected, accuracy declined with an increasing number of subjects in the dataset, reflecting the difficulty of multi-subject classification.

Electroencephalogram (EEG)-based identification offers a promising biometric solution by leveraging the uniqueness of individual brain activity patterns. This study proposes a framework based on a convolutional autoencoder (CAE) along with a traditional classifier for identifying individuals using EEG brainprints. The convolutional autoencoder extracts a compact and discriminative representation from the topographic data cubes that capture both spatial and temporal dynamics of neural oscillations. The latent tensor features extracted by the CAE are subsequently classified by a machine learning module utilizing Support Vector Machine (SVM), Random Forest (RF), k-Nearest Neighbor (KNN), and Artificial Neural Network (ANN) models. EEG data were collected under three conditions—resting state, music stimuli, and cognitive task—to investigate a diverse range of neural responses. Training and testing datasets were extracted from separate sessions to enable a true longitudinal analysis. The performance of the framework was evaluated using the Area Under the Curve (AUC) and accuracy (ACC) metrics. The effect of subject identifiability was also investigated. The proposed framework achieved a performance score up to a maximum AUC of 99.89% and ACC of 96.98%. These results demonstrate the effectiveness of the proposed automated subject-specific patterns in capturing stable EEG brainprints and support the potential of the proposed framework for reliable, session-independent EEG-based biometric identification.

Lung cancer is the leading cancer type that causes mortality in both men and women. Computer-aided detection (CAD) and diagnosis systems can play a very important role for helping physicians with cancer treatments. This study proposes a hierarchical deep-fusion learning scheme in a CAD framework for the detection of nodules from computed tomography (CT) scans. In the proposed hierarchical approach, a decision is made at each level individually employing the decisions from the previous level. Further, individual decisions are computed for several perspectives of a volume of interest. This study explores three different approaches to obtain decisions in a hierarchical fashion. The first model utilizes raw images. The second model uses a single type of feature image having salient content. The last model employs multi-type feature images. All models learn the parameters by means of supervised learning. The proposed CAD frameworks are tested using lung CT scans from the LIDC/IDRI database. The experimental results showed that the proposed multi-perspective hierarchical fusion approach significantly improves the performance of the classification. The proposed hierarchical deep-fusion learning model achieved a sensitivity of 95% with only 0.4 fp/scan.

Haptics is the study of coupling the human sense of touch and motion with computer systems. In this study, we analyze weight perception of the brain based on electroencephalographic signals, which is obtained using a pressure tactile as a stimulus. The state of weight perception is represented by the relative power of alpha and beta sub-bands of subjects. Independent Component Analysis is applied for post-processing. In the final stage, Linear Discriminant Analysis and Support Vector Machine techniques are utilized for classification of the brain signals as either holding a weight or being in a resting state. Our proposed method achieved satisfactory performance when Support Vector Machine with Radial Basis Function kernel classifier and relative power features, which are obtained from the alpha sub-band of four EEG channels, were used.

Music and Serious Games are separately useful alternative therapy methods for helping people with a cognitive disorder such as Attention Deficit Hyperactivity Disorder (ADHD). In this study, we aim to explore the effect of background music in game playing on children with and without ADHD. In addition, we propose an algorithm and a measure to quantify the activity of the brain. We designed Tetris, a well-known serious game, which plays Beethoven, Mozart, or no-music in the background. The brain activity of the subjects is recorded by means of electroencephalography (EEG) signals via a brain computer interface (BCI) device. We analyzed the effect of music utilizing time-domain features extracted from Alpha and Beta band of the EEG signals. The results showed that different music types have varying impact on the attention of children. In general, Mozart increases Beta-level brain activity while decreasing the Alpha-level one. On the other hand, Beethoven music increased concurrently both Alpha- and Beta-level activity for children with ADHD.

Facial action unit detection is an ongoing research area in image processing and computer vision. Different methods have been proposed in the recent years for accurate detection of action units in facial images for various views and lighting conditions. In this paper, we propose a new deep learning scheme that integrates convolutional neural networks with association rule mining. Convolutional neural network is used for automatic extraction of features from input images, and the results are exploited to build two different association rule mining-based classifiers. The obtained results show an improvement compared to a fully connected neural network, for which the features are originally designed, for some action units.