Explore Our Latest Research

 Andreas Miltiadous, Katerina D. Tzimourta, Vasileios Aspiotis, Theodora Afrantou, Markos G. Tsipouras, Nikolaos Giannakeas.

Description: The research focuses on the detection of Alzheimer's disease and frontotemporal dementia through electroencephalographic (EEG) signals. It involves a classification process starting with Independent Component Analysis (ICA) for pre-processing, followed by extraction of time, frequency and complexity features, feature selection based on significance and final classification with Gradient Boosting Decision Trees. The system achieved F1 scores: 92.27% for distinguishing dementia from healthy, 83.06% for Alzheimer's disease from healthy, and 80.67% for frontotemporal dementia from healthy.

Publication: published in the proceedings of the 2023 IEEE 36th International Symposium on Computer-Based Medical Systems (CBMS).

DOI: 10.1109/CBMS58004.2023.00335

Tzimourta, Nikolaos Giannakeas, Alexandros T. Alexandros Tzallas.

Description: This research presents DICE-net, an innovative Dual-Input Convolution Encoder network, for EEG signal classification in Alzheimer's disease (AD) patients. Using data from 36 AD patients, 23 with frontotemporal dementia (FTD), and 29 healthy subjects, the system integrates Convolution, Transformer Encoder, and Feed-Forward layers. DICE-net achieved 83.28% accuracy in distinguishing AD patients from healthy subjects (AD-CN) using Leave-One-Subject-Out validation, demonstrating improved generalizability over reference models. The study demonstrates the potential of exploiting the characteristics of EEG signals for early diagnosis and possible future extensions to frontotemporal dementia and other types.

Published in IEEE Access.

DOI: 10.1109/ACCESS.2023.3294618

Date of Publication:  12 July 2023  Publisher: IEEE

Andreas Miltiadous, Katerina D. Tzimourta, Nikolaos Giannakeas, Markos G. Tsipouras, Euripidis Glavas, Konstantinos Kalafatakis.

Description: This systematic review examines the use of machine learning algorithms for epilepsy detection based on published EEG databases. It presents signal processing and classification methodologies applied to different databases, focusing on the importance of automating diagnosis due to the time-consuming nature of manual analysis. From the analysis of 190 studies, we find an increased use of Convolutional Neural Networks (CNNs) that apply images through Time-Frequency Decomposition techniques.

Published in IEEE Access.

Publication date: 26 December 2022.

DOI: 10.1109/ACCESS.2022.3232563

Andreas Miltiadous, Vasileios Aspiotis, Konstantinos Sakkas, Nikolaos Giannakeas, Euripidis Glavas, Alexandros T. Tzallas.

This research proposes an experimental protocol to study stress in virtual environments (VR) using Electroencephalography (EEG). It combines data from EEG, Electrocardiography (ECG) and the Perceived Stress Scale to assess stress through a phobia induction scenario in VR. The protocol can be used to investigate functional brain connectivity and to develop machine learning algorithms for automatic classification of stress levels.

Published in 2022 7th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM).

Conference date: 23-25 September 2022.

Date added to IEEE Xplore: 1 November 2022.

DOI: 10.1109/SEEDA-CECNSM57760.2022.9932987

Panagiotis N. Smyrlis, Odysseas Tsakai, Konstantinos Vogklis, Nikolaos Giannakeas, Alexandros Tzallas, George F. Fragulis.

Abstracts: The YOLO architecture has shown excellent results in various detection and segmentation tasks, including biomedical image analysis. In this paper, we examine the effectiveness and applicability of the approach to cancer data through comparative tests with the state-of-the-art YOLOv8 algorithm. The experimental protocol includes a detailed analysis of different YOLOv8 approaches and provides an evaluative overview of the models through a plug-and-play application. The results present the performance characteristics of each model and examine the impact of the architecture combined with the development load and performance requirements.

Published in 2023 8th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM).

Conference date: 10-12 November 2023.

Date added to IEEE Xplore: 21 March 2024.

DOI: 10.1109/SEEDA-CECNSM61561.2023.10470849

Andreas Miltiadous, Katerina D. Tzimourta, Theodora Afrantou, Panagiotis Ioannidis, Nikolaos Grigoriadis, Dimitrios G. Tsalikakis, Pantelis Angelidis, Markos G. Tsipouras, Euripidis Glavas, Nikolaos Giannakeas, Alexandros T. Tzallas

Abstract:
There has been a growing interest in using the electroencephalogram (EEG) as a non-invasive diagnostic tool for neurodegenerative diseases. This article provides a detailed description of a resting-state EEG dataset from individuals with Alzheimer's disease, frontotemporal dementia, and healthy controls. The dataset was collected with a clinical EEG system using 19 scalp electrodes while participants were in a resting state with their eyes closed. It includes 36 Alzheimer's patients, 23 frontotemporal dementia patients, and 29 healthy age-matched controls. The data collection process included quality control measures to ensure data accuracy. The dataset is available in both raw and preprocessed forms in the standard BIDS format, with established methods like artifact subspace reconstruction and independent component analysis used for denoising. The dataset holds significant potential for future studies, particularly in the growing field of Alzheimer's EEG machine learning, and can aid in exploring brain activity alterations and connectivity in neurodegenerative diseases.

Published in: Data
Journal: 2023, 8(6), 95
DOI: 10.3390/data8060095

Victoria Zakopoulou, Elena Venizelou, Christos-Orestis Tsiantis, Alexandros Tzallas, George Dimakopoulos, Maria Syrrou

Abstract:
Developmental dyslexia (DD) is a multifactorial learning disorder influenced by various biological, neurophysiological, cognitive, and psychomotor factors. This study examines the association between early signs of DD and stress, focusing on polymorphisms/variants of HPA axis-related genes involved in stress regulation, and mitochondrial DNA copy number (mtDNAcn) as a sensitive stress biomarker. The study involved 314 children aged 5 to 6 years, including 20 preschoolers at risk for DD and 10 typically developing controls. The children underwent early identification screening and a 3-month intervention program. Genotyping of HPA axis gene variants and mtDNAcn estimation were conducted before and after the intervention. Multivariate analysis revealed significant differences in cognitive, psychomotor, and linguistic factors between the DD and control groups. Although no significant difference in mtDNAcn was found before the intervention, a statistically significant change was observed in the intervention group after treatment. The findings suggest that stress plays a critical role in the early onset of DD. Early diagnosis and intervention in preschool are essential for minimizing long-term negative effects and improving outcomes.

Published in: INPACT 2024 Conference Proceedings
DOI: 10.36315/2024inpact030

Andreas Miltiadous, Katerina D. Tzimourta, Nikolaos Giannakeas, Markos G. Tsipouras, Theodora Afrantou, Panagiotis Ioannidis, Alexandros T. Tzallas

Abstract:
Dementia is a clinical syndrome that involves progressive cognitive and emotional decline, severely affecting daily functioning. Alzheimer’s disease (AD) is the most prevalent neurodegenerative disorder, accounting for 50–70% of dementia cases. Frontotemporal dementia (FTD), another type of dementia, primarily affects personality and social skills due to degeneration of the prefrontal and anterior temporal cortex. This study explores the use of electroencephalogram (EEG) as an effective biomarker for identifying neuronal and cognitive dynamics in AD and FTD cases. Six supervised machine learning techniques were compared for classifying EEG signals from AD and FTD patients. Additionally, two validation methods, K-fold cross-validation and leave-one-patient-out cross-validation, were evaluated for their performance in this classification problem. The results showed that decision trees achieved an accuracy of 78.5% for AD detection, while random forests reached 86.3% accuracy for FTD detection.

Published in: Diagnostics
Date Published: 9 August 2021
DOI: 10.3390/diagnostics11081437

Vasileios Aspiotis, Dimitrios Peschos, Katerina D. Tzimourta, Markos G. Tsipouras, Al Husein Sami Abosaleh, Evangelos Antoniou

Abstract:
Touch is a key aspect of human interaction with the environment, influencing various developmental and cognitive processes. Recent studies have focused on understanding the electro-physiological brain activity related to haptic stimulation. This preliminary experiment aimed to classify EEG features from four healthy participants, who were asked to touch three different textures—smooth, rough, and water surface—using their fingertips. EEG recordings were collected and analyzed by extracting time- and frequency-based features, which were then fed into classifiers. The classification results showed a performance of 63% with the C4.5 algorithm and 76% with Random Forests using 10-fold cross-validation.

Published in: 2021 6th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM)
Date of Conference: 24-26 September 2021
Date Added to IEEE Xplore: 18 October 2021
DOI: 10.1109/SEEDA-CECNSM53056.2021.9566257

Vasileios Aspiotis, Andreas Miltiadous, Konstantinos Kalafatakis, Katerina D. Tzimourta, Nikolaos Giannakeas, Markos G. Tsipouras, Dimitrios Peschos, Euripidis Glavas, Alexandros T. Tzallas

Abstract:

Virtual reality (VR) advancements have made high-risk experimental scenarios, such as extreme height exposure, feasible for research. This study investigates stress system activation using VR-based high-altitude simulations combined with electroencephalography (EEG) and electrocardiography (ECG) sensors. Real-time monitoring of EEG and ECG biomarkers was performed while participants experienced the VR scenario. To process the movement-induced noise in EEG data, a robust signal preprocessing pipeline was employed. Statistical analysis revealed significant correlations between stress biomarkers and self-reported stress levels (via the Perceived Stress Scale questionnaire). Notably, occipital-region EEG biomarkers, such as beta and gamma band power and occipital alpha asymmetry, were associated with stress responses. Participants were divided into groups based on heart rate changes, further supporting biomarker differences linked to stress. These findings highlight EEG's potential as a reliable indicator of stress in VR-based experimental setups.

Published in: Sensors
Date Published: 3 August 2022
DOI: 10.3390/s22155792

Vasileios Christou, Andreas Miltiadous, Ioannis Tsoulos, Evaggelos Karvounis, Katerina D. Tzimourta, Markos G. Tsipouras, Nikolaos Anastasopoulos, Alexandros T. Tzallas, Nikolaos Giannakeas

Abstract:

Electroencephalography (EEG) is a key diagnostic tool for epilepsy, but the analysis of its intricate waveforms, including sporadic sharp waves and spikes indicative of epilepsy, remains challenging. This study examines how varying window sizes influence the accuracy of automated EEG signal classification using machine learning methods. Utilizing the University of Bonn's EEG dataset, EEG data were divided into overlapping epochs with window lengths ranging from 1 to 24 seconds. Statistical and spectral features were extracted for classification with four methods: a neural network trained via three different algorithms (Broyden–Fletcher–Goldfarb–Shanno, multistart method, genetic algorithm) and the k-nearest neighbors (k-NN) classifier. The results revealed that larger window sizes, approximately 21 seconds, significantly improved classification accuracy across the methods. This finding underscores the importance of selecting appropriate window sizes for enhancing the effectiveness of automated EEG epilepsy detection systems.

Published in: Sensors
Date Published: 27 November 2022
DOI: 10.3390/s22239233

Theodoros Lampros, Nikolaos Giannakeas, Konstantinos Kalafatakis, Markos Tsipouras, Alexandros Tzallas

Abstract:

The continuous monitoring of fetal heart activity during pregnancy is vital for identifying and preventing complications associated with fetal heart development. Noninvasive fetal electrocardiogram (NI-fECG) has become a significant focus due to its ability to provide prenatal diagnostic insights, such as beat-to-beat monitoring of the Fetal Heart Rate (FHR) and detailed electrophysiological analysis of fetal cardiac activity. However, extracting the fetal ECG from maternal abdominal recordings poses a challenge due to the interference from the maternal ECG, which is much stronger in amplitude.

This study introduces a novel hybrid methodology combining Reconstruction Independent Component Analysis (R-ICA) and Empirical Wavelet Transform (EWT) to accurately isolate the fetal ECG signal. The proposed RICA-EWT method was tested on real NI-fECG signals recorded from pregnant women at various stages of labor. The experimental results demonstrate its robustness and efficiency across different signal-to-noise ratio (SNR) levels, highlighting its potential for clinical applications in fetal heart monitoring.

Published in: Lecture Notes in Computer Science (Springer)
DOI: 10.1007/978-3-031-34171-7_3

Evangelos D. Spyrou, Christoforos Nestoris, Chrysostomos Stylios

Abstract:

Coprolalia, a hallmark symptom of Tourette’s syndrome, involves involuntary cursing that often targets societal taboos related to sexuality or social identity, leading to significant social isolation and heightened anxiety for patients. This behavior disrupts daily life activities and contributes to severe distress.

This research explores Coprolalia through the framework of bipartite graph matching, proposing that the brain reroutes ordinary thoughts, matching them with intrusive obscene counterparts. This decoding approach aims to elucidate the underlying neural mechanisms driving Coprolalia, potentially enabling patients to better understand and manage this behavior.

Additionally, the study introduces a game-theoretic model, inspired by the "Battle of the Sexes" paradigm, to analyze strategies for symptom minimization. The model presents two pure-strategy equilibria: blocking the intrusive thoughts or allowing them to pass. This mechanism supports patient resilience in social activities where symptoms are most pronounced and aligns with Cognitive Behavioral Therapy (CBT) principles. Specifically, CBT advocates for confronting intrusive thoughts with the goal of reducing their impact over time.

Published in: 2022 7th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM)
DOI: 10.1109/SEEDA-CECNSM57760.2022.9932906

Pavlos Christodoulides, Andreas Miltiadous, Katerina D. Tzimourta, Dimitrios Peschos, Georgios Ntritsos, Victoria Zakopoulou, Nikolaos Giannakeas, Loukas G. Astrakas, Markos G. Tsipouras, Konstantinos I. Tsamis, Euripidis Glavas, Alexandros T. Tzallas

Abstract:

The magnocellular pathway deficit theory has been proposed as a potential cause for dyslexia, suggesting that auditory and visual processing deficits are central to the disorder. Machine learning techniques applied to anatomical brain imaging have been used to classify these deficits, but EEG-based classification methods are increasingly recognized for their robustness and reliability. This study investigates the use of a Brain Computer Interface (BCI) device combined with an Interactive Linguistic Software Tool to classify dyslexia in a university population.

EEG signals were recorded from 12 university students with dyslexia and 14 age-matched typically developed individuals. The participants were assessed under three experimental conditions: auditory discrimination, visual recognition, and visual recognition with background music. Spectral features from the EEG rhythms (δ, θ, α, β, γ) were extracted and used to train a Random Forest classifier. The goal was to identify EEG features that distinguish dyslexia in different brain regions.

Results demonstrated high classification accuracy (above 95%) across the entire brain, with excellent performance in both hemispheres. The highest discrimination accuracy was observed in the third experimental condition, where background music was present. The findings suggest that combining BCI devices with linguistic tools can effectively classify dyslexia in higher education students and could be a valuable diagnostic approach.

Published in: Biomedical Signal Processing and Control
DOI: 10.1016/j.bspc.2022.103646

Alexandros Arjmand, Vasileios Christou, Ioannis G. Tsoulos, Markos G. Tsipouras, Alexandros T. Tzallas, Christos Gogos, Euripidis Glavas, Nikolaos Giannakeas

Abstract:
Non-alcoholic fatty liver disease (NAFLD) covers a range of chronic medical conditions varying from hepatocellular inflammation, which characterizes nonalcoholic steatohepatitis (NASH), to steatosis, the key element of nonalcoholic fatty liver (NAFL). This study presents an automated image analysis method for quantitatively assessing fat deposition in steatotic liver biopsy specimens. The method combines image processing, machine learning, and evolutionary algorithm optimization techniques, achieving a 1.93% mean classification error compared to semiquantitative interpretations by specialized hepatologists.

Published in: Array
DOI: 10.1016/j.array.2021.100078

Konstantinos Sakkas, Eirini Georgia Dimitriou, Niki Eleni Ntagka, Nikolaos Giannakeas, Konstantinos Kalafatakis

Abstract:
Parkinson’s disease is a neurological disorder characterized by motor and non-motor symptoms. Assessment methods, despite the many years of existence of the disease, lack individualized visualization. On the other hand, virtual reality promises immersion and realism. In this paper, we develop an integrated system for visualizing the gestures of Parkinson’s disease patients in a virtual reality environment. With this application, clinicians will have information about the unique motor patterns and challenges they must address in each individual patient’s case, while the collected data can travel and be easily and instantly visualized in any location. At the beginning of this research, the current terms of immersive technologies in conjunction with data visualization and Parkinson’s disease are described. Through an extensive systematic literature review, the technological developments in the field of Parkinson’s data visualization are presented. The findings of the review lead to the experimental procedure and implementation of the application. The conclusions drawn from this work fuel future extensions on the contribution of immersive technologies to various diseases.

Published in: Future Internet
DOI: 10.3390/fi16090305

Andreas Miltiadous, Vasileios Aspiotis, Dimitrios Peschos, Katerina D. Tzimourta, Al Husein Sami Abosaleh, Nikolaos Giannakeas, Alexandros Tzallas 

Abstract:
Touch sensation is a key modality that allows humans to understand and interact with their environment. More often than not, touch sensation depends on vision to accumulate and validate the received information. The ability to distinguish between materials and surfaces through active touch consists of a complex of neurophysiological operations. To unveil the functionality of these operations, neuroimaging and neurophysiological research tools are employed, with electroencephalography being the most used. In this paper, we attempt to distinguish between brain states when touching different natural textures (smooth, rough, and liquid). Recordings were obtained with a commercially available EEG wearable device. Time and frequency-based features were extracted, transformed with PCA decomposition, and an ensemble classifier combining Random Forest, Support Vector Machine, and Neural Network was utilized. High accuracy scores of 79.64% for the four-class problem and 89.34% for the three-class problem (Null-Rough-Water) were accordingly achieved. Thus, the methodology's robustness indicates its ability to classify different brain states under haptic stimuli.

Published in: Engineering, Technology & Applied Science Research
DOI: 10.48084/etasr.6455

Theodoros Lampros, Konstantinos Kalafatakis, Nikolaos Giannakeas, Markos G. Tsipouras, Euripidis Glavas, Alexandros T. Tzallas

Abstract:
Background and Objective: Electronic fetal heart monitoring is currently used during pregnancy throughout most of the developed world to detect risk conditions for both the mother and the fetus. Non-invasive fetal electrocardiogram (NI-fECG), recorded in the maternal abdomen, represents an alternative to cardiotocography, which could provide a more accurate estimate of fetal heart rate. Different methodologies, with varying advantages and disadvantages, have been developed for NI-fECG signal detection and processing.

Published in: Array
DOI: 10.1016/j.array.2023.100302