A Learnable and Explainable Wavelet Neural Network for EEG Artifacts Detection and Classification.

Electrical field of the heart (ECG) propagates throughout the body and introduce artifact in EEG recordings which may lead to incorrect interpretation of monitoring result. Hence in this paper, we present a method of automatic detection and reduction of ECG artifact from EEG. ECG has its own spike like property and periodicity. Moreover, it also has lack of correlation with the EEG signal. We have utilized the aforementioned properties to detect ECG artifact in EEG and have employed a method to remove it automatically. In the first step of the algorithm, an energy function based method is used to emphasize the R-waves of contaminated ECG artifact and thereafter, an adaptive thresholding method along with clustering is used to detect contaminated candidate R-spikes of ECG artifact in EEG signal. After that utilizing periodic information of R-wave, a searching mechanism is employed as post processing to detect the R-peaks more accurately. Thereafter, noise model of ECG artifact contaminated with EEG is generated and finally it is subtracted from the EEG recordings to decontaminate it from the artifact. Before subtraction, a time varying alignment procedure is applied to increase the effectiveness of the artifact reduction method. Results obtained from our extensive experiments show that the proposed method is effective and encouraging in terms of automatic ECG artifact detection and reduction from EEG signal.

The aim of this study was to develop and validate a deep learning-based algorithm (DLA) for automatic detection and grading of motion-related artifacts on arterial phase liver magnetic resonance imaging (MRI). Multistep DLA for detection and grading of motion-related artifacts, based on the modified ResNet-101 and U-net, were trained using 336 arterial phase images of gadoxetic acid-enhanced liver MRI examinations obtained in 2017 (training dataset; mean age, 68.6 years [range, 18-95]; 254 men). Motion-related artifacts were evaluated in 4 different MRI slices using a 3-tier grading system. In the validation dataset, 313 images from the same institution obtained in 2018 (internal validation dataset; mean age, 67.2 years [range, 21-87]; 228 men) and 329 from 3 different institutions (external validation dataset; mean age, 64.0 years [range, 23-90]; 214 men) were included, and the per-slice and per-examination performances for the detection of motion-related artifacts were evaluated. The per-slice sensitivity and specificity of the DLA for detecting grade 3 motion-related artifacts were 91.5% (97/106) and 96.8% (1134/1172) in the internal validation dataset and 93.3% (265/284) and 91.6% (948/1035) in the external validation dataset. The per-examination sensitivity and specificity were 92.0% (23/25) and 99.7% (287/288) in the internal validation dataset and 90.0% (72/80) and 96.0% (239/249) in the external validation dataset, respectively. The processing time of the DLA for automatic grading of motion-related artifacts was from 4.11 to 4.22 seconds per MRI examination. The DLA enabled automatic and instant detection and grading of motion-related artifacts on arterial phase gadoxetic acid-enhanced liver MRI.

Electroencephalography (EEG) is used in the diagnosis, monitoring, and prognostication of many neurological ailments including seizure, coma, sleep disorders, brain injury, and behavioral abnormalities. One of the primary challenges of EEG data is its sensitivity to a breadth of non-stationary noises caused by physiological-, movement-, and equipment-related artifacts. Existing solutions to artifact detection are deficient because they require experts to manually explore and annotate data for artifact segments. Existing solutions to artifact correction or removal are deficient because they assume that the incidence and specific characteristics of artifacts are similar across both subjects and tasks (i.e., “one-size-fits-all”). In this paper, we describe a novel EEG noise-reduction method that uses representation learning to perform patient- and task-specific artifact detection and correction. More specifically, our method extracts 58 clinically relevant features and applies an ensemble of unsupervised outlier detection algorithms to identify EEG artifacts that are unique to a given task and subject. The artifact segments are then passed to a deep encoder-decoder network for unsupervised artifact correction. We compared the performance of classification models trained with and without our method and observed a 10% relative improvement in performance when using our approach. Our method provides a flexible end-to-end unsupervised framework that can be applied to novel EEG data without the need for expert supervision and can be used for a variety of clinical decision tasks, including coma prognostication and degenerative illness detection. By making our method, code, and data publicly available, our work provides a tool that is of both immediate practical utility and may also serve as an important foundation for future efforts in this domain.

Our purpose in this study is to describe an algorithm for the automatic detection of linear artifacts in medical images. Linear artifacts arise as a result of many different forms of tissues and tissue boundaries within the imaging volume. Additionally, linear artifacts can arise for artificial structures such as radioactive seeds and radioactive linear sources. It is the purpose of the described algorithm to automatically detect linear artifacts of a certain length and diameter. The algorithm was written and compiled on a Pentium-4 based computer in the Microsoft Visual C/C++ language. Inert coils supplied by Radiomed Inc. were implanted into a standard prostate ultrasound phantom. Transaxial ultrasound images of the implanted phantom were obtained at 2 mm increments. The coded algorithm was then applied to the ultrasound imaging volume to automatically segment out the implanted coils. Thirteen coils were implanted in the prostate phantom. Thirteen coils were automatically identified in the imaging volume. An algorithm was developed to automatically determine the position and orientation of radioactive coils within an imaging volume. The algorithm successfully identified thirteen coils implanted in an ultrasound prostate phantom.

Objective: The activity of the brain can be recorded by means of an electroencephalogram (EEG). An EEG is a multichannel signal related to brain activity. However, EEG presents a wide variety of undesired artefacts. Removal of these artefacts is often done using blind source separation methods (BSS) and mainly those based on Independent Component Analysis (ICA). ICA-based methods are well-accepted in the literature for filtering artefacts and have proved to be satisfactory in most scenarios of interest. Our goal is to develop a generic and unsupervised ICA-based algorithm for EEG artefacts removal. Approach: The proposed algorithm makes use of a new unsupervised artefact detection, ICA and a statistical criterion to automatically select the artefact related independent components (ICs) requiring no human intervention. The algorithm is evaluated using both simulated and real EEG data with artefacts (SEEG and AEEG). A comparison between the proposed unsupervised selection of ICs related to the artefact and other supervised selection is also presented. Main results: A new unsupervised ICA-based algorithm to filter artefacts, where ICs related to each artefact are automatically selected. It can be used in online applications, it preserves most of the original information among the artefacts and removes different types of artefacts. Significance: ICA-based methods for filtering artefacts prevail in the literature. The work in this article is important insofar as it addresses the problem of automatic selection of ICs in ICA-based methods. The selection is unsupervised, avoiding the manual ICs selection or a learning process involved in other methods. Our method is a generic algorithm that allows removing EEG artefacts of various types and, unlike some ICA-based algorithms, it retains most of the original information among the artefacts. Within the algorithm, the artefact detection method implemented does not require human intervention either.

The National Sleep Research Resource (NSRR, www.sleepdata.org) features thousands of polysomnograms (PSGs) that can be analyzed for further understanding how variations in physiological signals associate with health outcomes. Quantitative EEG analysis may help characterize physiological variation. However, analysis of large datasets collected in uncontrolled settings requires a robust pipeline including artifact detectors. To promote community-wide use of PSG data, we developed an open-source, automated pipeline for spectral analysis of sleep EEGs and tested the level of agreement with traditional analysis. We used data from the C3-A2 EEG lead in a sample of PSGs from 161 women participating in the Study of Osteoporotic Fractures. The traditional approach used manual artifact removal on 4-s basis and application of commercial spectral analysis software. Automated analysis included spectral power-based artifact detection on 30-s basis and generation of summary figures for adjudication. We compared automatic and manual artifact detection epoch-by-epoch and then compared the average EEG spectral power density in six frequency bands obtained with the two approaches using correlation analysis, Bland-Altman plots and Wilcoxon test. The automated artifact detection algorithm had high specificity (96.8 to 99.4% in NREM, 96.9 to 99.1% in REM depending on the criterion for comparing 4-s with 30-s epochs) but lower sensitivity (26.7 to 38.1% in NREM, 9.1 to 27.4% in REM). However, we found no clinically or statistically significant differences in power density values, and results were highly correlated (Spearman’s r>0.99). Large artifacts (total power >99th percentile) were removed with sensitivity up to 90.9% in NREM, 87.7% in REM, specificity 96.6% and 96.9%. The automated pipeline generated similar results to those obtained with standard approach, while reducing analysis time 100-fold. This Matlab toolset, publicly available on the NSRR website, can be used to analyze thousands of recordings, allowing for its application in genetics and epidemiological research. The work presented in this paper was funded by: NIH R24 HL114473, 1R01HL083075-01, R01HL098433, R01 HL098433-02S1, 1U34HL105277-01, 1R01HL110068-01A1 1R01HL113338-01, R21 HL108226, P20 NS076965, R01 HL109493 and R03MH108908; and a research agreement with the Emma B. Bradley Hospital/Brown University.

Electroencephalographic signals have been shown to be extremely susceptible to noise from both external and internal sources. In this way, electroencephalography (EEG) can be considered a tool for analyzing the electrical activity of the brain. In the literature, there are numerous methods for detecting artifacts in EEG signals. Nevertheless, these approaches not attain a high level of accuracy in artifact detection. New techniques are needed to increase the accuracy level. Predictive Preventive and Personalized Medicine (PPPM) is becoming the focus of healthcare initiatives to guarantee the early identification of various types of artifacts. In this research, a binary classification is developed as physiological and non-physiological artifact detection. Various techniques are targeted for artifact classification as deep learning (DL) which is the most efficient strategy used for image and signals classification. Based on the convolutional neural network (CNN) model, we have classified the various artifact types. The Nihon Kohden device was used to acquire the available EEG data, which is collected from 32 channels. Our approach is validated with a CNN architecture for binary artifact classification. With a maximum accuracy of 99.55%, the proposed methodology delivers powerful classification findings.

Probe-based confocal laser endomicroscopy (pCLE) is a subcellular in vivo imaging technique capable of producing images that enable diagnosis of malign structural modifications in epithelial tissue. Images acquired with pCLE are, however, often tainted by significant artifacts that impair diagnosis. This is especially detrimental for automated image analysis, which is why said images are often excluded from recognition pipelines. We present an approach for the automatic detection of motion artifacts in pCLE images and apply this methodology to a data set of 15 thousand images of epithelial tissue acquired in the oral cavity and the vocal folds. The approach is based on transfer learning from intermediate endpoints within a pre-trained Inception v3 network with tailored preprocessing. For detection within the non-rectangular pCLE images, we perform pooling within the activation maps of the network and evaluate this at different network depths. We achieved area under the ROC curve values of 0.92 with the proposed method, compared to 0.80 for the best feature-based machine learning approach. Our overall accuracy with the presented approach is 94.8%. Over traditional machine learning approaches with state-of-the-art features, we achieved significantly improved overall performance.

We apply and compare various Artificial Neural Network (ANN) and other algorithms for automatic morphological classification of galaxies. The ANNs are presented here mathematically, as non-linear extensions of conventional statistical methods in Astronomy. The methods are illustrated using different subsets Artificial Neural Network (ANN) and other algorithms for automatic morphological classification of galaxies. The ANNs are presented here mathematically, as non-linear extensions of conventional statistical methods in Astronomy. The methods are illustrated using different subsets from the ESO-LV catalogue, for which both machine parameters and human classification are available. The main methods we explore are: (i) Principal Component Analysis (PCA) which tells how independent and informative the input parameters are. (ii) Encoder Neural Network which allows us to find both linear (PCA-like) and non-linear combinations of the input, illustrating an example of unsupervised ANN. (iii) Supervised ANN (using the Backpropagation or Quasi-Newton algorithms) based on a training set for which the human classification is known. Here the output for previously unclassified galaxies can be interpreted as either a continuous (analog) output (e.g. $T$-type) or a Bayesian {\it a posteriori} probability for each class. Although the ESO-LV parameters are sub-optimal, the success of the ANN in reproducing the human classification is 2 $T$-type units, similar to the degree of agreement between two human experts who classify the same galaxy images on plate material. We also examine the aspects of ANN configurations, reproducibility, scaling of input parameters and redshift information.

Given the non-Euclidean topology inherent in electroencephalogram (EEG) electrode configurations, graph-based approaches, particularly graph neural networks, have shown notable success across diverse EEG classification tasks. However, since the cerebral cortex lobes function individually and/or collaboratively across diverse tasks, there exist substantial differences between intra-lobe and inter-lobe brain intrinsic functional connectivity. Existing graph networks for EEG classification are based on homogeneous graphs, yet the nature of the cerebral cortex aligns more closely with a heterogeneous graph structure. To this end, we propose HeteroEEG for EEG classification, which to the best of our knowledge is the first to reframe the challenge of exploring EEG spatial information, especially decoupling different types of brain lobes and functional connections, as heterogeneous graph reasoning. Specifically, HeteroEEG is designed to be a dual-branch network aware of spatial, spectral, and temporal EEG features. Experimental results justify the superiority of HeteroEEG in pain and emotion recognition compared with other state-of-the-art studies. The heterogeneous graph construction of HeteroEEG may shed light on future graph-based EEG classification network design.

Cervical cancer is one of the primary factors that endanger women’s health, and Thinprep cytologic test (TCT) is the common testing tool for the early diagnosis of cervical cancer. However, it is tedious and time-consuming for pathologists to assess and find abnormal cells in many TCT samples. Thus, automatic detection of abnormal cervical cells is highly demanded. Nevertheless, false positive cells are inevitable after automatic detection. It is still a burden for the pathologist if the false positive rate is high. To this end, here we propose a semi-supervised cervical cell diagnosis method that can significantly reduce the false positive rate. First, we incorporate a detection network to localize the suspicious abnormal cervical cells. Then, we design a semi-supervised classification network to identify whether the cervical cells are truly abnormal or not. To boost the performance of the semi-supervised classification network, and make full use of the localizing information derived from the detection network, we use the predicted bounding boxes of the detection network as an additional constraint for the attention masks from the classification network. Besides, we also develop a novel consistency constraint between the teacher and student models to guarantee the robustness of the network. Our experimental results show that our network can achieve satisfactory classification accuracy using only a limited number of labeled cells, and also greatly reduce the false positive rate in cervical cell detection.

The driver's cognitive and physiological states affect his/her ability to control the vehicle. Thus, these driver states are essential to the safety of automobiles. The design of advanced driver assistance systems (ADAS) or autonomous vehicles will depend on their ability to interact effectively with the driver. A deeper understanding of the driver state is, therefore, paramount. Electroencephalography (EEG) is proven to be one of the most effective methods for driver state monitoring and human error detection. This paper discusses EEG-based driver state detection systems and their corresponding analysis algorithms over the last three decades. First, the commonly used EEG system setup for driver state studies is introduced. Then, the EEG signal preprocessing, feature extraction, and classification algorithms for driver state detection are reviewed. Finally, EEG-based driver state monitoring research is reviewed in-depth, and its future development is discussed. It is concluded that the current EEG-based driver state monitoring algorithms are promising for safety applications. However, many improvements are still required in EEG artifact reduction, real-time processing, and between-subject classification accuracy.

Background: Recorded brain activity analysis is a major tool used in modern medicine and sleep-based research. However, the volume and complexity of the data make manual investigation difficult. Furthermore, high inter-subject variability presents a challenge to deep learning techniques, especially in the detection of sleep-related EEG patterns and artifacts. Methods: Sleep Pattern Identifier (SPI), a hybrid classifier, was used in this study to address the aforementioned issues. SPI tests the efficacy of artifact detection while examining strategies and tactics used in real-world sleep research. A focus was on inter-subject variability, especially in data gathered from participants suffering from sleep disorders. For comparison and visual inspection of the data, formal statistical measures like accuracy, model loss, precision, recall, and ROC were employed. Results: With an accuracy of 94.85%, the suggested model outperformed current techniques and demonstrated higher accuracy. Further, the Sleep-EDF dataset was used in this investigation. Conclusion: The conclusion highlights the effectiveness of the SPI hybrid classifier in identifying EEG patterns, especially in situations where sleep research is conducted in the real world. Research methodologies based on sleep have made significant progress in handling artifacts and adapting to inter-subject variability.

One of the biggest problems in EEG recordings is the contamination of the signal caused by artifacts because these interferences hinder the analysis of real neural information. Thus, their elimination while preserving as much brain data as possible is a key procedure before the study of the EEG. To address the automatic removal of craniofacial artifacts, this paper proposes a two-stages procedure: the former one is the detection stage - where both a MLP neural network and a dynamic threshold method are applied to detect the contaminated areas of the EEG-, while the latter is the removal stage - combining CCA and EEMD algorithms to remove the artifact data only. Experimental results show that both detection methods are comparable, but with the dynamic threshold detection slightly outperforming the MLP. Also, the combined technique can completely remove those artifacts scattered in all the EEG channels. This study will be extended to ocular artifacts, where more complex models would be required.

Parkinson's disease (PD) is characterized by widespread disruptions in neural oscillations and network dynamics, which can be captured through resting-state EEG biomarkers. Galvanic vestibular stimulation (GVS) has emerged as a promising noninvasive neuromodulation technique to modulate these neural patterns. However, EEG recordings during GVS are severely contaminated by highamplitude stimulation artifacts, especially when exploring a wide range of stimulation protocols. In this study, we designed a data acquisition protocol involving 304 distinct subthreshold GVS waveforms, each with a unique temporal profile, to investigate their effects on brain activity. These stimuli induced strong artifacts in the EEG signal, particularly during the stimulation interval. To recover clean EEG signals, we developed a multi-stage preprocessing pipeline combining regression-based artifact suppression, canonical correlation analysis (CCA), and independent component analysis (ICA), supported by machine learning classifiers for automatic detection and removal of GVS, EOG, and EMG artifacts. We evaluated the effectiveness of this pipeline through classification of EEG signals from PD patients and healthy controls across three temporal segments: pre-stimulation (Prestim), stimulation (Stim), and post-stimulation (Post-stim). Despite the intense artifacts in the Stim interval, classification accuracy reached 82.46%, closely matching the performance in Pre-stim (85.06%) and Post-stim (91.67%) intervals. This confirms that the artifact removal process successfully preserved disease-relevant neural information even during active stimulation. Beyond classification, we conducted additional evaluations including temporal consistency analysis of biomarkers, correlation of model coefficients across intervals, and visual inspection of signal quality. These assessments demonstrated that the cleaned EEG signals retained physiologically meaningful patterns and stable biomarker profiles across time. Our findings show that EEG signals recorded during GVS can be reliably cleaned and analyzed, enabling rapid screening of stimulation protocols and paving the way for personalized neuromodulation strategies in Parkinson's disease.