Marios Antonakakis

Prof. Marios Antonakakis

School of Electronic Engineering, Hellenic Mediterranean University, Heraklion, Greece
orcid

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 December 2020 in NeuroImage
Verified
Journal Article
Open access
Marios AntonakakisMarios Antonakakis
Marios Antonakakis
SSSophie Schrader
Sophie Schrader
ÜAÜmit Aydin
Ümit Aydin
AKAsad Khan
Asad Khan
JGJoachim Gross
Joachim Gross
Carsten WoltersCarsten Wolters
Carsten Wolters
Skull conductivity has a substantial influence on EEG and combined EEG and MEG source analysis as well as on optimized transcranial electric stimulation. To overcome the use of standard literature values, we propose a non-invasive two-level calibration procedure to estimate skull conductivity individually in a group study with twenty healthy adults. Our procedure requires only an additional run of combined somatosensory evoked potential and field data, which can be easily integrated in EEG/MEG experiments. The calibration procedure uses the P20/N20 topographies and subject-specific realistic head models from MRI. We investigate the inter-subject variability of skull conductivity and relate it to skull thickness, age and gender of the subjects, to the individual scalp P20/N20 surface distance between the P20 potential peak and the N20 potential trough as well as to the individual source depth of the P20/N20 source. We found a considerable inter-subject variability for (calibrated) skull conductivity (8.44 ± 4.84 mS/m) and skull thickness (5.97 ± 1.19 mm) with a statistically significant correlation between them (rho = 0.52). Age showed a statistically significant negative correlation with skull conductivity (rho = -0.5). Furthermore, P20/N20 surface distance and source depth showed large inter-subject variability of 12.08 ± 3.21 cm and 15.45 ± 4.54 mm, respectively, but there was no significant correlation between them. We also found no significant differences among gender subgroups for the investigated measures. It is thus important to take the inter-subject variability of skull conductivity and thickness into account by means of using subject-specific calibrated realistic head modeling.

1 April 2016 in International Journal of Psychophysiology
Verified
Journal Article
Closed access
Marios AntonakakisMarios Antonakakis
Marios Antonakakis
SDStavros I. Dimitriadis
Stavros I. Dimitriadis
MZMichalis Zervakis
Michalis Zervakis
SMSifis Micheloyannis
Sifis Micheloyannis
RRRoozbeh Rezaie
Roozbeh Rezaie
APAndrew C. Papanicolaou
Andrew C. Papanicolaou

9 August 2019 in Human Brain Mapping
Verified
Journal Article
Open access
PDF file
Marios AntonakakisMarios Antonakakis
Marios Antonakakis
SSSophie Schrader
Sophie Schrader
AWAndreas Wollbrink
Andreas Wollbrink
RORobert Oostenveld
Robert Oostenveld
SRStefan Rampp
Stefan Rampp
CWCarsten H. Wolters
Carsten H. Wolters
Modeling and experimental parameters influence the Electro‐ (EEG) and Magnetoencephalography (MEG) source analysis of the somatosensory P20/N20 component. In a sensitivity group study, we compare P20/N20 source analysis due to different stimulation type (Electric‐Wrist [EW], Braille‐Tactile [BT], or Pneumato‐Tactile [PT]), measurement modality (combined EEG/MEG – EMEG, EEG, or MEG) and head model (standard or individually skull‐conductivity calibrated including brain anisotropic conductivity). Considerable differences between pairs of stimulation types occurred (EW‐BT: 8.7 ± 3.3 mm/27.1° ± 16.4°, BT‐PT: 9 ± 5 mm/29.9° ± 17.3°, and EW‐PT: 9.8 ± 7.4 mm/15.9° ± 16.5° and 75% strength reduction of BT or PT when compared to EW) regardless of the head model used. EMEG has nearly no localization differences to MEG, but large ones to EEG (16.1 ± 4.9 mm), while source orientation differences are non‐negligible to both EEG (14° ± 3.7°) and MEG (12.5° ± 10.9°). Our calibration results show a considerable inter‐subject variability (3.1–14 mS/m) for skull conductivity. The comparison due to different head model show localization differences smaller for EMEG (EW: 3.4 ± 2.4 mm, BT: 3.7 ± 3.4 mm, and PT: 5.9 ± 6.8 mm) than for EEG (EW: 8.6 ± 8.3 mm, BT: 11.8 ± 6.2 mm, and PT: 10.5 ± 5.3 mm), while source orientation differences for EMEG (EW: 15.4° ± 6.3°, BT: 25.7° ± 15.2° and PT: 14° ± 11.5°) and EEG (EW: 14.6° ± 9.5°, BT: 16.3° ± 11.1° and PT: 12.9° ± 8.9°) are in the same range. Our results show that stimulation type, modality and head modeling all have a non‐negligible influence on the source reconstruction of the P20/N20 component. The complementary information of both modalities in EMEG can be exploited on the basis of detailed and individualized head models.

15 January 2020 in Frontiers in Computational Neuroscience
Verified
Journal Article
Open access
Marios AntonakakisMarios Antonakakis
Marios Antonakakis
SDStavros I. Dimitriadis
Stavros I. Dimitriadis
MZMichalis Zervakis
Michalis Zervakis
APAndrew C. Papanicolaou
Andrew C. Papanicolaou
GZGeorge Zouridakis
George Zouridakis
Dynamic Functional Connectivity (DFC) analysis is a promising approach for the characterization of brain electrophysiological activity. In this study, we investigated abnormal alterations due to mild Traumatic Brain Injury (mTBI) using DFC of the source reconstructed magnetoencephalographic (MEG) resting-state recordings. Brain activity in several well-known frequency bands was first reconstructed using beamforming of the MEG data to determine ninety anatomical brain regions of interest. A DFC graph was formulated using the imaginary part of phase-locking values, which were obtained from 30 mTBI patients and 50 healthy controls (HC). Subsequently, we estimated normalized Laplacian transformations of individual, statistically and topologically filtered quasi-static graphs. The corresponding eigenvalues of each node synchronization were then computed and through the neural-gas algorithm, we quantized the evolution of the eigenvalues resulting in distinct network microstates (NMstates). The discrimination level between the two groups was assessed using an iterative cross-validation classification scheme with features either the NMstates in each frequency band, or the combination of the so-called chronnectomics (flexibility index, occupancy time of NMstate, and Dwell time) with the complexity index over the evolution of the NMstates across all frequency bands. Classification performance based on chronnectomics showed 80% accuracy, 99% sensitivity, and 49% specificity. However, performance was much higher (accuracy: 91 - 97 %, sensitivity: 100%, and specificity: 77 – 93 %) when focusing on the microstates. Exploring the mean node degree within and between brain anatomical networks (default mode network, frontoparietal, occipital, cingulo-opercular, and sensorimotor), a reduced pattern occurred from lower to higher frequency bands, with statistically significant stronger degrees for the HC than the mTBI group. A higher entropic profile on the temporal evolution of the modularity index was observed for both NMstates for the mTBI group across frequencies. A significant difference in the flexibility index was observed between the two groups for the β frequency band. The latter finding may support a central role of the thalamus impairment in mTBI. The current study considers a complete set of frequency-dependent connectomic markers of mTBI-caused alterations in brain connectivity that potentially could serve as markers to assess the return of an injured subject back to normality.

KSkoushlendra Kumar Singh
koushlendra Kumar Singh
SKSuraj Kumar
Suraj Kumar
Marios AntonakakisMarios Antonakakis
Marios Antonakakis
KMKonstantia Moirogiorgou
Konstantia Moirogiorgou
ADAnirudh Deep
Anirudh Deep
MZMichails Zervakis
Michails Zervakis
Breast cancer is the most common cancer in women worldwide. It is the most frequently diagnosed cancer among women in 140 countries out of 184 reporting countries. Lesions of breast cancer are abnormal areas in the breast tissues. Various types of breast cancer lesions include (1) microcalcifications, (2) masses, (3) architectural distortion, and (4) bilateral asymmetry. Microcalcification can be classified as benign, malignant, and benign without a callback. In the present manuscript, we propose an automatic pipeline for the detection of various categories of microcalcification. We performed deep learning using convolution neural networks (CNNs) for the automatic detection and classification of all three categories of microcalcification. CNN was applied using four different optimizers (ADAM, ADAGrad, ADADelta, and RMSProp). The input images of a size of 299 × 299 × 3, with fully connected RELU and SoftMax output activation functions, were utilized in this study. The feature map was obtained using the pretrained InceptionResNetV2 model. The performance evaluation of our classification scheme was tested on a curated breast imaging subset of the DDSM mammogram dataset (CBIS–DDSM), and the results were expressed in terms of sensitivity, specificity, accuracy, and area under the curve (AUC). Our proposed classification scheme outperforms the ability of previously used deep learning approaches and classical machine learning schemes.

AKApostolos Karasmanoglou
Apostolos Karasmanoglou
Marios AntonakakisMarios Antonakakis
Marios Antonakakis
MZMichails Zervakis
Michails Zervakis
Epilepsy is one of the most common brain diseases, characterized by frequent recurrent seizures or “ictal” states. A patient experiences uncontrollable muscular contractions, inducing loss of mobility and balance, which may result in injury or even death during these ictal states. Extensive investigation is vital to establish a systematic approach for predicting and informing patients about oncoming seizures ahead of time. Most methodologies developed are focused on the detection of abnormalities using mostly electroencephalogram (EEG) recordings. In this regard, research has indicated that certain pre-ictal alterations in the Autonomic Nervous System (ANS) can be detected in patient electrocardiogram (ECG) signals. The latter could potentially provide the basis for a robust seizure prediction approach. The recently proposed ECG-based seizure warning systems utilize machine learning models to classify a patient’s condition. Such approaches require the incorporation of large, diverse, and thoroughly annotated ECG datasets, limiting their application potential. In this work, we investigate anomaly detection models in a patient-specific context with low supervision requirements. Specifically, we consider One-Class SVM (OCSVM), Minimum Covariance Determinant (MCD) Estimator, and Local Outlier Factor (LOF) models to quantify the novelty or abnormality of pre-ictal short-term (2–3 min) Heart Rate Variability (HRV) features of patients, trained on a reference interval considered to contain stable heart rate as the only form of supervision. Our models are evaluated against labels that were either hand-picked or automatically generated (weak labels) by a two-phase clustering procedure for samples of the “Post-Ictal Heart Rate Oscillations in Partial Epilepsy” (PIHROPE) dataset recorded by the Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, achieving detection in 9 out of 10 cases, with average AUCs of over 93% across all models and warning times ranging from 6 to 30 min prior to seizure. The proposed anomaly detection and monitoring approach can potentially pave the way for early detection and warning of seizure incidents based on body sensor inputs.