Multiscale neuro-inspired models for interpretation of EEG signals in patients with epilepsy.

OBJECTIVE: The aim is to gain insight into the pathophysiological mechanisms underlying interictal epileptiform discharges observed in electroencephalographic (EEG) and stereo-EEG (SEEG, depth electrodes) recordings performed during pre-surgical evaluation of patients with drug-resistant epilepsy. METHODS: We developed novel neuro-inspired computational models of the human cerebral cortex at three different levels of description: i) microscale (detailed neuron models), ii) mesoscale (neuronal mass models) and iii) macroscale (whole brain models). Although conceptually different, micro- and mesoscale models share some similar features, such as the typology of neurons (pyramidal cells and three types of interneurons), their spatial arrangement in cortical layers, and their synaptic connectivity (excitatory and inhibitory). The whole brain model consists of a large-scale network of interconnected neuronal masses, with connectivity based on the human connectome. RESULTS: For these three levels of description, the fine-tuning of free parameters and the quantitative comparison with real data allowed us to reproduce interictal epileptiform discharges with a high degree of fidelity and to formulate hypotheses about the cell- and network-related mechanisms underlying the generation of fast ripples and SEEG-recorded epileptic spikes and spike-waves. CONCLUSIONS: The proposed models provide valuable insights into the pathophysiological mechanisms underlying the generation of epileptic events. The knowledge gained from these models effectively complements the clinical analysis of SEEG data collected during the evaluation of patients with epilepsy. SIGNIFICANCE: These models are likely to play a key role in the mechanistic interpretation of epileptiform activity.

Electrophysiological brain imaging based on simulation-driven deep learning in the context of epilepsy.

Identifying the location, the spatial extent and the electrical activity of distributed brain sources in the context of epilepsy through ElectroEncephaloGraphy (EEG) recordings is a challenging task because of the highly ill-posed nature of the underlying Electrophysiological Source Imaging (ESI) problem. To guarantee a unique solution, most existing ESI methods pay more attention to solve this inverse problem by imposing physiological constraints. This paper proposes an efficient ESI approach based on simulation-driven deep learning. Epileptic High-resolution 256-channels scalp EEG (Hr-EEG) signals are simulated in a realistic manner to train the proposed patient-specific model. More particularly, a computational neural mass model developed in our team is used to generate the temporal dynamics of the activity of each dipole while the forward problem is solved using a patient-specific three-shell realistic head model and the boundary element method. A Temporal Convolutional Network (TCN) is considered in the proposed model to capture local spatial patterns. To enable the model to observe the EEG signals from different scale levels, the multi-scale strategy is leveraged to capture the overall features and fine-grain features by adjusting the convolutional kernel size. Then, the Long Short-Term Memory (LSTM) is used to extract temporal dependencies among the computed spatial features. The performance of the proposed method is evaluated through three different scenarios of realistic synthetic interictal Hr-EEG data as well as on real interictal Hr-EEG data acquired in three patients with drug-resistant partial epilepsy, during their presurgical evaluation. A performance comparison study is also conducted with two other deep learning-based methods and four classical ESI techniques. The proposed model achieved a Dipole Localization Error (DLE) of 1.39 and Normalized Hamming Distance (NHD) of 0.28 in the case of one patch with SNR of 10 dB. In the case of two uncorrelated patches with an SNR of 10 dB, obtained DLE and NHD were respectively 1.50 and 0.28. Even in the more challenging scenario of two correlated patches with an SNR of 10 dB, the proposed approach still achieved a DLE of 3.74 and an NHD of 0.43. The results obtained on simulated data demonstrate that the proposed method outperforms the existing methods for different signal-to-noise and source configurations. The good behavior of the proposed method is also confirmed on real interictal EEG data. The robustness with respect to noise makes it a promising and alternative tool to localize epileptic brain areas and to reconstruct their electrical activities from EEG signals.

Signal processing and computational modeling for interpretation of SEEG-recorded interictal epileptiform discharges in epileptogenic and non-epileptogenic zones.

Objective.In partial epilepsies, interictal epileptiform discharges (IEDs) are paroxysmal events observed in epileptogenic zone (EZ) and non-epileptogenic zone (NEZ). IEDs' generation and recurrence are subject to different hypotheses: they appear through glutamatergic and gamma-aminobutyric acidergic (GABAergic) processes; they may trigger seizures or prevent seizure propagation. This paper focuses on a specific class of IEDs, spike-waves (SWs), characterized by a short-duration spike followed by a longer duration wave, both of the same polarity. Signal analysis and neurophysiological mathematical models are used to interpret puzzling IED generation.Approach.Interictal activity was recorded by intracranial stereo-electroencephalography (SEEG) electrodes in five different patients. SEEG experts identified the epileptic and non-epileptic zones in which IEDs were detected. After quantifying spatial and temporal features of the detected IEDs, the most significant features for classifying epileptic and non-epileptic zones were determined. A neurophysiologically-plausible mathematical model was then introduced to simulate the IEDs and understand the underlying differences observed in epileptic and non-epileptic zone IEDs.Main results.Two classes of SWs were identified according to subtle differences in morphology and timing of the spike and wave component. Results showed that type-1 SWs were generated in epileptogenic regions also involved at seizure onset, while type-2 SWs were produced in the propagation or non-involved areas. The modeling study indicated that synaptic kinetics, cortical organization, and network interactions determined the morphology of the simulated SEEG signals. Modeling results suggested that the IED morphologies were linked to the degree of preserved inhibition.Significance.This work contributes to the understanding of different mechanisms generating IEDs in epileptic networks. The combination of signal analysis and computational models provides an efficient framework for exploring IEDs in partial epilepsies and classifying EZ and NEZ.

Selective enhancement of low-gamma activity by tACS improves phonemic processing and reading accuracy in dyslexia.

The phonological deficit in dyslexia is associated with altered low-gamma oscillatory function in left auditory cortex, but a causal relationship between oscillatory function and phonemic processing has never been established. After confirming a deficit at 30 Hz with electroencephalography (EEG), we applied 20 minutes of transcranial alternating current stimulation (tACS) to transiently restore this activity in adults with dyslexia. The intervention significantly improved phonological processing and reading accuracy as measured immediately after tACS. The effect occurred selectively for a 30-Hz stimulation in the dyslexia group. Importantly, we observed that the focal intervention over the left auditory cortex also decreased 30-Hz activity in the right superior temporal cortex, resulting in reinstating a left dominance for the oscillatory response. These findings establish a causal role of neural oscillations in phonological processing and offer solid neurophysiological grounds for a potential correction of low-gamma anomalies and for alleviating the phonological deficit in dyslexia.

Modelling acute and lasting effects of tDCS on epileptic activity.

Transcranial Direct brain stimulation (tDCS) is commonly used in order to modulate cortical networks activity during physiological processes through the application of weak electrical fields with scalp electrodes. Cathodal stimulation has been shown to decrease brain excitability in the context of epilepsy, with variable success. However, the cellular mechanisms responsible for the acute and the long-lasting effect of tDCS remain elusive. Using a novel approach of computational modeling that combines detailed but functionally integrated neurons we built a physiologically-based thalamocortical column. This model comprises 10,000 individual neurons made of pyramidal cells, and 3 types of gamma-aminobutyric acid (GABA) -ergic cells (VIP, PV, and SST) respecting the anatomy, layers, projection, connectivity and neurites orientation. Simulating realistic electric fields in term of intensity, main results showed that 1) tDCS effects are best explained by modulation of the presynaptic probability of release 2) tDCS affects the dynamic of cortical network only if a sufficient number of neurons are modulated 3)VIP GABAergic interneurons of the superficial layer of the cortex are especially affected by tDCS 4) Long lasting effect depends on glutamatergic synaptic plasticity.

COALIA: A Computational Model of Human EEG for Consciousness Research.

Understanding the origin of the main physiological processes involved in consciousness is a major challenge of contemporary neuroscience, with crucial implications for the study of Disorders of Consciousness (DOC). The difficulties in achieving this task include the considerable quantity of experimental data in this field, along with the non-intuitive, nonlinear nature of neuronal dynamics. One possibility of integrating the main results from the experimental literature into a cohesive framework, while accounting for nonlinear brain dynamics, is the use of physiologically-inspired computational models. In this study, we present a physiologically-grounded computational model, attempting to account for the main micro-circuits identified in the human cortex, while including the specificities of each neuronal type. More specifically, the model accounts for thalamo-cortical (vertical) regulation of cortico-cortical (horizontal) connectivity, which is a central mechanism for brain information integration and processing. The distinct neuronal assemblies communicate through feedforward and feedback excitatory and inhibitory synaptic connections implemented in a template brain accounting for long-range connectome. The EEG generated by this physiologically-based simulated brain is validated through comparison with brain rhythms recorded in humans in two states of consciousness (wakefulness, sleep). Using the model, it is possible to reproduce the local disynaptic disinhibition of basket cells (fast GABAergic inhibition) and glutamatergic pyramidal neurons through long-range activation of vasoactive intestinal-peptide (VIP) interneurons that induced inhibition of somatostatin positive (SST) interneurons. The model (COALIA) predicts that the strength and dynamics of the thalamic output on the cortex control the local and long-range cortical processing of information. Furthermore, the model reproduces and explains clinical results regarding the complexity of transcranial magnetic stimulation TMS-evoked EEG responses in DOC patients and healthy volunteers, through a modulation of thalamo-cortical connectivity that governs the level of cortico-cortical communication. This new model provides a quantitative framework to accelerate the study of the physiological mechanisms involved in the emergence, maintenance and disruption (sleep, anesthesia, DOC) of consciousness.

On the origin of epileptic High Frequency Oscillations observed on clinical electrodes.

OBJECTIVE: In this study we aim to identify the key (patho)physiological mechanisms and biophysical factors which impact the observability and spectral features of High Frequency Oscillations (HFOs). METHODS: In order to accurately replicate HFOs we developed virtual-brain/virtual-electrode simulation environment combining novel neurophysiological models of neuronal populations with biophysical models for the source/sensor relationship. Both (patho)physiological mechanisms (synaptic transmission, depolarizing GABAA effect, hyperexcitability) and physical factors (geometry of extended cortical sources, size and position of electrodes) were taken into account. Simulated HFOs were compared to real HFOs extracted from intracerebral recordings of 2 patients. RESULTS: Our results revealed that HFO pathological activity is being generated by feed-forward activation of cortical interneurons that produce fast depolarizing GABAergic post-synaptic potentials (PSPs) onto pyramidal cells. Out of phase patterns of depolarizing GABAergic PSPs explained the shape, entropy and spatiotemporal features of real human HFOs. CONCLUSIONS: The terminology "high-frequency oscillation" (HFO) might be misleading as the fast ripple component (200-600 Hz) is more likely a "high-frequency activity" (HFA), the origin of which is independent from any oscillatory process. SIGNIFICANCE: New insights regarding the origins and observability of HFOs along depth-EEG electrodes were gained in terms of spatial extent and 3D geometry of neuronal sources.

Spatial projection as a preprocessing step for EEG source reconstruction using spatiotemporal Kalman filtering.

The reconstruction of brain sources from non-invasive electroencephalography (EEG) or magnetoencephalography (MEG) via source imaging can be distorted by information redundancy in case of high-resolution recordings. Dimensionality reduction approaches such as spatial projection may be used to alleviate this problem. In this proof-of-principle paper we apply spatial projection to solve the problem of information redundancy in case of source reconstruction via spatiotemporal Kalman filtering (STKF), which is based on state-space modeling. We compare two approaches for incorporating spatial projection into the STKF algorithm and select the best approach based on its performance in source localization with respect to accurate estimation of source location, lack of spurious sources, computational speed and small number of required optimization steps in state-space model parameter estimation. We use state-of-the-art simulated EEG data based on neuronal population models, for which the number and location of sources is known, to validate the source reconstruction results of the STKF. The incorporation of spatial projection into the STKF algorithm solved the problem of information redundancy, resulting in correct source localization with no spurious sources, and decreased the overall computational time in STKF analysis. The results help make STKF analyses of high-density EEG, MEG or simultaneous MEG-EEG data more feasible.

Source reconstruction via the spatiotemporal Kalman filter and LORETA from EEG time series with 32 or fewer electrodes.

The clinical routine of non-invasive electroencephalography (EEG) is usually performed with 8-40 electrodes, especially in long-term monitoring, infants or emergency care. There is a need in clinical and scientific brain imaging to develop inverse solution methods that can reconstruct brain sources from these low-density EEG recordings. In this proof-of-principle paper we investigate the performance of the spatiotemporal Kalman filter (STKF) in EEG source reconstruction with 9-, 19- and 32- electrodes. We used simulated EEG data of epileptic spikes generated from lateral frontal and lateral temporal brain sources using state-of-the-art neuronal population models. For validation of source reconstruction, we compared STKF results to the location of the simulated source and to the results of low-resolution brain electromagnetic tomography (LORETA) standard inverse solution. STKF consistently showed less localization bias compared to LORETA, especially when the number of electrodes was decreased. The results encourage further research into the application of the STKF in source reconstruction of brain activity from low-density EEG recordings.

Automatic Detection and Classification of High-Frequency Oscillations in Depth-EEG Signals.

GOAL: Interictal high-frequency oscillations (HFOs [30-600 Hz]) have proven to be relevant biomarkers in epilepsy. In this paper, four categories of HFOs are considered: Gamma ([30-80 Hz]), high-gamma ([80-120 Hz]), ripples ([120-250 Hz]), and fast-ripples ([250-600 Hz]). A universal detector of the four types of HFOs is proposed. It has the advantages of 1) classifying HFOs, and thus, being robust to inter and intrasubject variability; 2) rejecting artefacts, thus being specific. METHODS: Gabor atoms are tuned to cover the physiological bands. Gabor transform is then used to detect HFOs in intracerebral electroencephalography (iEEG) signals recorded in patients candidate to epilepsy surgery. To extract relevant features, energy ratios, along with event duration, are investigated. Discriminant ratios are optimized so as to maximize among the four types of HFOs and artefacts. A multiclass support vector machine (SVM) is used to classify detected events. Pseudoreal signals are simulated to measure the performance of the method when the ground truth is known. RESULTS: Experiments are conducted on simulated and on human iEEG signals. The proposed method shows high performance in terms of sensitivity and false discovery rate. CONCLUSION: The methods have the advantages of detecting and discriminating all types of HFOs as well as avoiding false detections caused by artefacts. SIGNIFICANCE: Experimental results show the feasibility of a robust and universal detector.

Identification of Interictal Epileptic Networks from Dense-EEG.

Epilepsy is a network disease. The epileptic network usually involves spatially distributed brain regions. In this context, noninvasive M/EEG source connectivity is an emerging technique to identify functional brain networks at cortical level from noninvasive recordings. In this paper, we analyze the effect of the two key factors involved in EEG source connectivity processing: (i) the algorithm used in the solution of the EEG inverse problem and (ii) the method used in the estimation of the functional connectivity. We evaluate four inverse solutions algorithms (dSPM, wMNE, sLORETA and cMEM) and four connectivity measures (r 2, h 2, PLV, and MI) on data simulated from a combined biophysical/physiological model to generate realistic interictal epileptic spikes reflected in scalp EEG. We use a new network-based similarity index to compare between the network identified by each of the inverse/connectivity combination and the original network generated in the model. The method will be also applied on real data recorded from one epileptic patient who underwent a full presurgical evaluation for drug-resistant focal epilepsy. In simulated data, results revealed that the selection of the inverse/connectivity combination has a significant impact on the identified networks. Results suggested that nonlinear methods (nonlinear correlation coefficient, phase synchronization and mutual information) for measuring the connectivity are more efficient than the linear one (the cross correlation coefficient). The wMNE inverse solution showed higher performance than dSPM, cMEM and sLORETA. In real data, the combination (wMNE/PLV) led to a very good matching between the interictal epileptic network identified from noninvasive EEG recordings and the network obtained from connectivity analysis of intracerebral EEG recordings. These results suggest that source connectivity method, when appropriately configured, is able to extract highly relevant diagnostic information about networks involved in interictal epileptic spikes from non-invasive dense-EEG data.

A Penalized Semialgebraic Deflation ICA Algorithm for the Efficient Extraction of Interictal Epileptic Signals.

As a noninvasive technique, electroencephalography (EEG) is commonly used to monitor the brain signals of patients with epilepsy such as the interictal epileptic spikes. However, the recorded data are often corrupted by artifacts originating, for example, from muscle activities, which may have much higher amplitudes than the interictal epileptic signals of interest. To remove these artifacts, a number of independent component analysis (ICA) techniques were successfully applied. In this paper, we propose a new deflation ICA algorithm, called penalized semialgebraic unitary deflation (P-SAUD) algorithm, that improves upon classical ICA methods by leading to a considerably reduced computational complexity at equivalent performance. This is achieved by employing a penalized semialgebraic extraction scheme, which permits us to identify the epileptic components of interest (interictal spikes) first and obviates the need of extracting subsequent components. The proposed method is evaluated on physiologically plausible simulated EEG data and actual measurements of three patients. The results are compared to those of several popular ICA algorithms as well as second-order blind source separation methods, demonstrating that P-SAUD extracts the epileptic spikes with the same accuracy as the best ICA methods, but reduces the computational complexity by a factor of 10 for 32-channel recordings. This superior computational efficiency is of particular interest considering the increasing use of high-resolution EEG recordings, whose analysis requires algorithms with low computational cost.

Denoising of Ictal EEG Data Using Semi-Blind Source Separation Methods Based on Time-Frequency Priors.

Removing muscle activity from ictal ElectroEncephaloGram (EEG) data is an essential preprocessing step in diagnosis and study of epileptic disorders. Indeed, at the very beginning of seizures, ictal EEG has a low amplitude and its morphology in the time domain is quite similar to muscular activity. Contrary to the time domain, ictal signals have specific characteristics in the time-frequency domain. In this paper, we use the time-frequency signature of ictal discharges as a priori information on the sources of interest. To extract the time-frequency signature of ictal sources, we use the Canonical Correlation Analysis (CCA) method. Then, we propose two time-frequency based semi-blind source separation approaches, namely the Time-Frequency-Generalized EigenValue Decomposition (TF-GEVD) and the Time-Frequency-Denoising Source Separation (TF-DSS), for the denoising of ictal signals based on these time-frequency signatures. The performance of the proposed methods is compared with that of CCA and Independent Component Analysis (ICA) approaches for the denoising of simulated ictal EEGs and of real ictal data. The results show the superiority of the proposed methods in comparison with CCA and ICA.

Classification of high frequency oscillations in epileptic intracerebral EEG.

High Frequency Oscillations (HFOs 40-500 Hz), recorded from intracerebral electroencephalography (iEEG) in epileptic patients, are categorized into four distinct sub-bands (Gamma, High-Gamma, Ripples and Fast Ripples). They have recently been used as a reliable biomarker of epileptogenic zones. The objective of this paper is to investigate the possibility of discriminating between the different classes of HFOs which physiological/pathological value is critical for diagnostic but remains to be clarified. The proposed method is based on the definition of a relevant feature vector built from energy ratios (computed using Wavelet Transform-WT) in a-priori-defined frequency bands. It makes use of a multiclass Linear Discriminant Analysis (LDA) and is applied to iEEG signals recorded in patients candidate to epilepsy surgery. Results obtained from bootstrap on training/test datasets indicate high performances in terms of sensitivity and specificity.

EEG source connectivity analysis: from dense array recordings to brain networks.

The recent past years have seen a noticeable increase of interest for electroencephalography (EEG) to analyze functional connectivity through brain sources reconstructed from scalp signals. Although considerable advances have been done both on the recording and analysis of EEG signals, a number of methodological questions are still open regarding the optimal way to process the data in order to identify brain networks. In this paper, we analyze the impact of three factors that intervene in this processing: i) the number of scalp electrodes, ii) the combination between the algorithm used to solve the EEG inverse problem and the algorithm used to measure the functional connectivity and iii) the frequency bands retained to estimate the functional connectivity among neocortical sources. Using High-Resolution (hr) EEG recordings in healthy volunteers, we evaluated these factors on evoked responses during picture recognition and naming task. The main reason for selection this task is that a solid literature background is available about involved brain networks (ground truth). From this a priori information, we propose a performance criterion based on the number of connections identified in the regions of interest (ROI) that belong to potentially activated networks. Our results show that the three studied factors have a dramatic impact on the final result (the identified network in the source space) as strong discrepancies were evidenced depending on the methods used. They also suggest that the combination of weighted Minimum Norm Estimator (wMNE) and the Phase Synchronization (PS) methods applied on High-Resolution EEG in beta/gamma bands provides the best performance in term of topological distance between the identified network and the expected network in the above-mentioned cognitive task.

From oscillatory transcranial current stimulation to scalp EEG changes: a biophysical and physiological modeling study.

Both biophysical and neurophysiological aspects need to be considered to assess the impact of electric fields induced by transcranial current stimulation (tCS) on the cerebral cortex and the subsequent effects occurring on scalp EEG. The objective of this work was to elaborate a global model allowing for the simulation of scalp EEG signals under tCS. In our integrated modeling approach, realistic meshes of the head tissues and of the stimulation electrodes were first built to map the generated electric field distribution on the cortical surface. Secondly, source activities at various cortical macro-regions were generated by means of a computational model of neuronal populations. The model parameters were adjusted so that populations generated an oscillating activity around 10 Hz resembling typical EEG alpha activity. In order to account for tCS effects and following current biophysical models, the calculated component of the electric field normal to the cortex was used to locally influence the activity of neuronal populations. Lastly, EEG under both spontaneous and tACS-stimulated (transcranial sinunoidal tCS from 4 to 16 Hz) brain activity was simulated at the level of scalp electrodes by solving the forward problem in the aforementioned realistic head model. Under the 10 Hz-tACS condition, a significant increase in alpha power occurred in simulated scalp EEG signals as compared to the no-stimulation condition. This increase involved most channels bilaterally, was more pronounced on posterior electrodes and was only significant for tACS frequencies from 8 to 12 Hz. The immediate effects of tACS in the model agreed with the post-tACS results previously reported in real subjects. Moreover, additional information was also brought by the model at other electrode positions or stimulation frequency. This suggests that our modeling approach can be used to compare, interpret and predict changes occurring on EEG with respect to parameters used in specific stimulation configurations.

Effects of transcranial Direct Current Stimulation (tDCS) on cortical activity: a computational modeling study.

Although it is well-admitted that transcranial Direct Current Stimulation (tDCS) allows for interacting with brain endogenous rhythms, the exact mechanisms by which externally-applied fields modulate the activity of neurons remain elusive. In this study a novel computational model (a neural mass model including subpopulations of pyramidal cells and inhibitory interneurons mediating synaptic currents with either slow or fast kinetics) of the cerebral cortex was elaborated to investigate the local effects of tDCS on neuronal populations based on an in-vivo experimental study. Model parameters were adjusted to reproduce evoked potentials (EPs) recorded from the somatosensory cortex of the rabbit in response to air-puffs applied on the whiskers. EPs were simulated under control condition (no tDCS) as well as under anodal and cathodal tDCS fields. Results first revealed that a feed-forward inhibition mechanism must be included in the model for accurate simulation of actual EPs (peaks and latencies). Interestingly, results revealed that externally-applied fields are also likely to affect interneurons. Indeed, when interneurons get polarized then the characteristics of simulated EPs become closer to those of real EPs. In particular, under anodal tDCS condition, more realistic EPs could be obtained when pyramidal cells were depolarized and, simultaneously, slow (resp. fast) interneurons became de- (resp. hyper-) polarized. Geometrical characteristics of interneurons might provide some explanations for this effect.

Transcranial current brain stimulation (tCS): models and technologies.

In this paper, we provide a broad overview of models and technologies pertaining to transcranial current brain stimulation (tCS), a family of related noninvasive techniques including direct current (tDCS), alternating current (tACS), and random noise current stimulation (tRNS). These techniques are based on the delivery of weak currents through the scalp (with electrode current intensity to area ratios of about 0.3-5 A/m2) at low frequencies (typically < 1 kHz) resulting in weak electric fields in the brain (with amplitudes of about 0.2-2 V/m). Here we review the biophysics and simulation of noninvasive, current-controlled generation of electric fields in the human brain and the models for the interaction of these electric fields with neurons, including a survey of in vitro and in vivo related studies. Finally, we outline directions for future fundamental and technological research.

[Cognitive-behavioral specialized units: predictive factors of readmissions within three months]. / Unité cognitivo-comportementale : facteurs prédictifs de réadmission à 3 mois.

BACKGROUND: Behavioral and psychological symptoms of dementia (BPSD) are frequent and belong to the natural evolution of the disease. Specialized cognitive-behavioral units (Unités cognitivo-comportementales) were created, in France (plan Alzheimer 2008-2012), to cope with this problem. Despite a stay in such a unit, some patients have to be rehospitalized. The main aim of the current study was to highlight the predictive factors of readmissions. METHOD: Descriptive, retrospective study of demented patients ≥75 years, hospitalized between January 2010 and April 2011. We compared patients that had to be rehospitalized within 3 months (group 1), with the patients that did not need to be rehospitalized or after 3 months of time (group 2). Patients characteristics included: basic daily living activities (French GIR score), MMSE score, neuropsychiatric inventory score, type of BPSD, length of stay and antipsychotropic drugs. RESULTS: Two hundred thirty-five patients were included including, 147 women (62.5%), with mean age of 82.74±7.13 years. SPCD was the main reason for hospitalization. Thirty patients (12.77%) belonged to group 1. The mean number of psychotropic treatments increased during the stay (p=0.02), particularly in group 2 (p=0.01). The NPI score decreased during the hospitalization in both groups. Linear regression analysis showed that behavioral type of symptoms (OR: 3.18; 95% CI 1.32-7.65) and association of antidepressant and antipsychotic drugs (OR: 4.77; 95% CI 1.35-16.83) were significantly predictive of an early readmission. The risk of readmission also significantly decreased as the length of stay increased. CONCLUSION: This work confirms the specificity and the need for such units. The results will help improving the outcome of demented patients with BPSD and treated with different antipsychotropic drugs.