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.

A reduced complexity ECG imaging model for regularized inversion optimization.

The resolution of the inverse problem of electrocardiography represents a major interest in the diagnosis and catheter-based therapy of cardiac arrhythmia. In this context, the ability to simulate several cardiac electrical behaviors was crucial for evaluating and comparing the performance of inversion methods. For this application, existing models are either too complex or do not produce realistic cardiac patterns. In this work, a low-resolution heart-torso model generating realistic whole heart cardiac mappings and electrocardiograms in healthy and pathological cases is designed. This model was built upon a simplified heart-torso geometry and implements the monodomain formalism by using the finite element method. In addition, a model reduction step through a sensitivity analysis was proposed where parameters were identified using an evolutionary optimization approach. Finally, the study illustrates the usefulness of the proposed model by comparing the performance of different variants of Tikhonov-based inversion methods for the determination of the regularization parameter in healthy, ischemic and ventricular tachycardia scenarios. First, results of the sensitivity analysis show that among 58 parameters only 25 are influent. Note also that the level of influence of the parameters depends on the heart region. Besides, the synthesized electrocardiograms globally present the same characteristic shape compared to the reference once with a correlation value that reaches 88%. Regarding inverse problem, results highlight that only Robust Generalized Cross Validation and Discrepancy Principle provide best performance, with a quasi-perfect success rate for both, and a respective relative error, between the generated electrocardiograms to the reference one, of 0.75 and 0.62.

Epileptic Seizure Prediction Using Deep Neural Networks Via Transfer Learning and Multi-Feature Fusion.

Epilepsy is one of the most common neurological diseases, which can seriously affect the patient's psychological well-being and quality of life. An accurate and reliable seizure prediction system can generate alarm before epileptic seizures to provide patients and their caregivers with sufficient time to take appropriate action. This study proposes an efficient seizure prediction system based on deep learning in order to anticipate the onset of the seizure as early as possible. Handcrafted features extracted based on the prior knowledge and hidden deep features are complementarily fused through the feature fusion module, and then the hybrid features are fed into the multiplicative long short-term memory (MLSTM) to explore the temporal dependency in EEG signals. A one-dimensional channel attention mechanism is implemented to emphasize the more representative information in the multi-channel output of the MLSTM. Finally, a transfer learning strategy is proposed to transfer the weights of the base model trained on the EEG data of all patients to the target patient model, and the latter is then continuously trained using the EEG data of the target patient. The proposed method achieves an average sensitivity of 95.56% and a false positive rate (FPR) of 0.27/h on the SWEC-ETHZ intracranial EEG data. For the more challenging CHB-MIT scalp EEG database, an average sensitivity of 89.47% and a FPR of 0.34/h are obtained. Experimental results demonstrate that the proposed method has good robustness and generalization ability in both intracranial and scalp EEG signals.

Localization of extended brain sources from EEG/MEG: the ExSo-MUSIC approach.

We propose a new MUSIC-like method, called 2q-ExSo-MUSIC (q ≥ 1). This method is an extension of the 2q-MUSIC (q ≥ 1) approach for solving the EEG/MEG inverse problem, when spatially-extended neocortical sources ("ExSo") are considered. It introduces a novel ExSo-MUSIC principle. The novelty is two-fold: i) the parameterization of the spatial source distribution that leads to an appropriate metric in the context of distributed brain sources and ii) the introduction of an original, efficient and low-cost way of optimizing this metric. In 2q-ExSo-MUSIC, the possible use of higher order statistics (q ≥ 2) offers a better robustness with respect to Gaussian noise of unknown spatial coherence and modeling errors. As a result we reduced the penalizing effects of both the background cerebral activity that can be seen as a Gaussian and spatially correlated noise, and the modeling errors induced by the non-exact resolution of the forward problem. Computer results on simulated EEG signals obtained with physiologically-relevant models of both the sources and the volume conductor show a highly increased performance of our 2q-ExSo-MUSIC method as compared to the classical 2q-MUSIC algorithms.

Turning Tangent Empirical Mode Decomposition: A Framework for Mono- and Multivariate Signals.

A novel Empirical Mode Decomposition (EMD) algorithm, called 2T-EMD, for both mono- and multivariate signals is proposed in this paper. It differs from the other approaches by its computational lightness and its algorithmic simplicity. The method is essentially based on a redefinition of the signal mean envelope, computed thanks to new characteristic points, which offers the possibility to decompose multivariate signals without any projection. The scope of application of the novel algorithm is specified, and a comparison of the 2T-EMD technique with classical methods is performed on various simulated mono- and multivariate signals. The monovariate behaviour of the proposed method on noisy signals is then validated by decomposing a fractional Gaussian noise and an application to real life EEG data is finally presented.

Independent component analysis for rectal bleeding prediction following prostate cancer radiotherapy.

BACKGROUND AND PURPOSE: To evaluate the benefit of independent component analysis (ICA)-based models for predicting rectal bleeding (RB) following prostate cancer radiotherapy. MATERIALS AND METHODS: A total of 593 irradiated prostate cancer patients were prospectively analyzed for Grade ≥2 RB. ICA was used to extract two informative subspaces (presenting RB or not) from the rectal DVHs, enabling a set of new pICA parameters to be estimated. These DVH-based parameters, along with others from the principal component analysis (PCA) and functional PCA, were compared to "standard" features (patient/treatment characteristics and DVH bins) using the Cox proportional hazards model for RB prediction. The whole cohort was divided into: (i) training (N = 339) for ICA-based subspace identification and Cox regression model identification and (ii) validation (N = 254) for RB prediction capability evaluation using the C-index and the area under the receiving operating curve (AUC), by comparing predicted and observed toxicity probabilities. RESULTS: In the training cohort, multivariate Cox analysis retained pICA and PC as significant parameters of RB with 0.65 C-index. For the validation cohort, the C-index increased from 0.64 when pICA was not included in the Cox model to 0.78 when including pICA parameters. When pICA was not included, the AUC for 3-, 5-, and 8-year RB prediction were 0.68, 0.66, and 0.64, respectively. When included, the AUC increased to 0.83, 0.80, and 0.78, respectively. CONCLUSION: Among the many various extracted or calculated features, ICA parameters improved RB prediction following prostate cancer radiotherapy.

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.

A new parameter computed with independent component analysis to predict rectal toxicity following prostate cancer radiotherapy.

The main challenge in prostate cancer radiotherapy is to deliver the prescribed dose to the clinical target while minimizing the dose to the neighboring organs at risk and thus avoiding subsequent toxicity-related events. With the aim of improving toxicity prediction following prostate cancer radiotherapy, the goal of our work is to propose a new predictive variable computed with independent component analysis to predict late rectal toxicity, and to compare its performance to other models (logistic regression, normal tissue complication probability model and recent principal component analysis approach). Clinical data and dose-volume histograms were collected from 216 patients having received 3D conformal radiation for prostate cancer with at least two years of follow-up. Independent component analysis was trained to predict the risk of 3-year rectal bleeding Grade ≥ 2. The performance of all the models was assessed by computing the area under the receiving operating characteristic curve. Clinical parameters combined with the new variable were found to be predictors of rectal bleeding. The mean area under the receiving operating curve for our proposed approach was 0:75. The AUC values for the logistic regression, the Lyman-Kutcher-Burman model and the recent principal component analysis approach were 0:62, 0:53 and 0:62, respectively. Our proposed new variable may be an useful new tool in predicting late rectal toxicity. It appears as a strong predictive variable to improve classical models.

On feature extraction and classification in prostate cancer radiotherapy using tensor decompositions.

External beam radiotherapy is commonly prescribed for prostate cancer. Although new radiation techniques allow high doses to be delivered to the target, the surrounding healthy organs (rectum and bladder) may suffer from irradiation, which might produce undesirable side-effects. Hence, the understanding of the complex toxicity dose-volume effect relationships is crucial to adapt the treatment, thereby decreasing the risk of toxicity. In this paper, we introduce a novel method to classify patients at risk of presenting rectal bleeding based on a Deterministic Multi-way Analysis (DMA) of three-dimensional planned dose distributions across a population. After a non-rigid spatial alignment of the anatomies applied to the dose distributions, the proposed method seeks for two bases of vectors representing bleeding and non bleeding patients by using the Canonical Polyadic (CP) decomposition of two fourth order arrays of the planned doses. A patient is then classified according to its distance to the subspaces spanned by both bases. A total of 99 patients treated for prostate cancer were used to analyze and test the performance of the proposed approach, named CP-DMA, in a leave-one-out cross validation scheme. Results were compared with supervised (linear discriminant analysis, support vector machine, K-means, K-nearest neighbor) and unsupervised (recent principal component analysis-based algorithm, and multidimensional classification method) approaches based on the registered dose distribution. Moreover, CP-DMA was also compared with the Normal Tissue Complication Probability (NTCP) model. The CP-DMA method allowed rectal bleeding patients to be classified with good specificity and sensitivity values, outperforming the classical approaches.

A Novel Classification Method for Prediction of Rectal Bleeding in Prostate Cancer Radiotherapy Based on a Semi-Nonnegative ICA of 3D Planned Dose Distributions.

The understanding of dose/side-effects relationships in prostate cancer radiotherapy is crucial to define appropriate individual's constraints for the therapy planning. Most of the existing methods to predict side-effects do not fully exploit the rich spatial information conveyed by the three-dimensional planned dose distributions. We propose a new classification method for three-dimensional individuals' doses, based on a new semi-nonnegative ICA algorithm to identify patients at risk of presenting rectal bleeding from a population treated for prostate cancer. The method first determines two bases of vectors from the population data: the two bases span vector subspaces, which characterize patients with and without rectal bleeding, respectively. The classification is then achieved by calculating the distance of a given patient to the two subspaces. The results, obtained on a cohort of 87 patients (at two year follow-up) treated with radiotherapy, showed high performance in terms of sensitivity and specificity.

Automatic detection of fast ripples.

OBJECTIVE: We propose a new method for automatic detection of fast ripples (FRs) which have been identified as a potential biomarker of epileptogenic processes. METHODS: This method is based on a two-stage procedure: (i) global detection of events of interest (EOIs, defined as transient signals accompanied with an energy increase in the frequency band of interest 250-600Hz) and (ii) local energy vs. frequency analysis of detected EOIs for classification as FRs, interictal epileptic spikes or artifacts. For this second stage, two variants were implemented based either on Fourier or wavelet transform. The method was evaluated on simulated and real depth-EEG signals (human, animal). The performance criterion was based on receiving operator characteristics. RESULTS: The proposed detector showed high performance in terms of sensitivity and specificity. CONCLUSIONS: As designed to specifically detect FRs, the method outperforms any method simply based on the detection of energy changes in high-pass filtered signals and avoids spurious detections caused by sharp transient events often present in raw signals. SIGNIFICANCE: In most of epilepsy surgery units, huge data sets are generated during pre-surgical evaluation. We think that the proposed detection method can dramatically decrease the workload in assessing the presence of FRs in intracranial EEGs.

On joint diagonalization of cumulant matrices for independent component analysis of MRS and EEG signals.

An extension of the original implementation of JADE, named eJADE((1)) hereafter, was proposed in 2001 to perform independent component analysis for any combination of statistical orders greater than or equal to three. More precisely, eJADE((1)) relies on the joint diagonalization of a set of several cumulant matrices corresponding to different matrix slices of one or several higher order cumulant tensors. An efficient way, without lose of statistical information, of reducing the number of third and fourth order cumulant matrices to be jointly diagonalized is proposed in this paper. The resulting approach, named eJADE(3,4)((2)), can be interpreted as an improvement of the eJADE(3,4)((1)) method. A performance comparison with classical methods is conducted in the context of MRS and EEG signals showing the good behavior of our technique.

Brain source localization using a fourth-order deflation scheme.

In this paper, a high-resolution method for solving potentially ill-posed inverse problems is proposed. This method named FO-D-MUSIC allows for localization of brain current sources with unconstrained orientations from surface electroencephalographic (EEG) or magnetoencephalographic (MEG) data using spherical or realistic head geometries. The FO-D-MUSIC method is based on the following: 1) the separability of the data transfer matrix as a function of location and orientation parameters, 2) the fourth-order (FO) virtual array theory, and 3) the deflation concept extended to FO statistics accounting for the presence of potentially but not completely statistically dependent sources. Computer results display the superiority of the FO-D-MUSIC approach in different situations (very closed sources, small number of electrodes, additive Gaussian noise with unknown spatial covariance, etc.) compared to classical algorithms.

Blind source separation methods applied to synthesized polysomnographic recordings: a comparative study.

Blind Source Separation (BSS) problems, under the assumption of static mixture, were extensively explored from the theoretical point of view. Powerful algorithms are now at hand to deal with many concrete BSS applications. Nevertheless, the performances of BSS methods, for a given biomedical application, are rarely investigated. The aim of this paper is to perform quantitative comparisons between various well-known BSS techniques. To do so, synthetic data, reproducing real polysomnographic recordings, are considered.

Fourth order approaches for localization of brain current sources.

Two high resolution methods solving inverse problems potentially ill-posed, named 4-MUSIC and 4-RapMUSIC, are proposed. They allow for localization of brain current sources with unconstrained orientations from surface electro-or magneto-encephalographic data using spherical or realistic head geometries. The 4-MUSIC and 4-RapMUSIC methods are based on i) the separability of the data transfer matrix as a function of location and orientation parameters and ii) the fourth order (FO) virtual array theory. In addition, 4-RapMUSIC uses the deflation concept extended to FO statistics accounting for the presence of potentially but not totally coherent sources. Computer results display the superiority of the 4-RapMUSIC approach in different situations (two closed sources, additive Gaussian noise with unknown spatial covariance, ...) especially over classical algorithms.