Single-subject classification of schizophrenia using event-related potentials obtained during auditory and visual oddball paradigms

A systematic review of EEG based automated schizophrenia classification through machine learning and deep learning

The electroencephalogram (EEG) serves as an essential tool in exploring brain activity and holds particular importance in the field of mental health research. This review paper examines the application of artificial intelligence (AI), encompassing machine learning (ML) and deep learning (DL), for classifying schizophrenia (SCZ) through EEG. It includes a thorough literature review that addresses the difficulties, methodologies, and discoveries in this field. ML approaches utilize conventional models like Support Vector Machines and Decision Trees, which are interpretable and effective with smaller data sets. In contrast, DL techniques, which use neural networks such as convolutional neural networks (CNNs) and long short-term memory networks (LSTMs), are more adaptable to intricate EEG patterns but require significant data and computational power. Both ML and DL face challenges concerning data quality and ethical issues. This paper underscores the importance of integrating various techniques to enhance schizophrenia diagnosis and highlights AI's potential role in this process. It also acknowledges the necessity for collaborative and ethically informed approaches in the automated classification of SCZ using AI.

Sampling inequalities affect generalization of neuroimaging-based diagnostic classifiers in psychiatry

BACKGROUND

The development of machine learning models for aiding in the diagnosis of mental disorder is recognized as a significant breakthrough in the field of psychiatry. However, clinical practice of such models remains a challenge, with poor generalizability being a major limitation.

METHODS

Here, we conducted a pre-registered meta-research assessment on neuroimaging-based models in the psychiatric literature, quantitatively examining global and regional sampling issues over recent decades, from a view that has been relatively underexplored. A total of 476 studies (n = 118,137) were included in the current assessment. Based on these findings, we built a comprehensive 5-star rating system to quantitatively evaluate the quality of existing machine learning models for psychiatric diagnoses.

RESULTS

A global sampling inequality in these models was revealed quantitatively (sampling Gini coefficient (G) = 0.81, p < .01), varying across different countries (regions) (e.g., China, G = 0.47; the USA, G = 0.58; Germany, G = 0.78; the UK, G = 0.87). Furthermore, the severity of this sampling inequality was significantly predicted by national economic levels (β =  - 2.75, p < .001, R²_adj = 0.40; r =  - .84, 95% CI: - .41 to - .97), and was plausibly predictable for model performance, with higher sampling inequality for reporting higher classification accuracy. Further analyses showed that lack of independent testing (84.24% of models, 95% CI: 81.0-87.5%), improper cross-validation (51.68% of models, 95% CI: 47.2-56.2%), and poor technical transparency (87.8% of models, 95% CI: 84.9-90.8%)/availability (80.88% of models, 95% CI: 77.3-84.4%) are prevailing in current diagnostic classifiers despite improvements over time. Relating to these observations, model performances were found decreased in studies with independent cross-country sampling validations (all p < .001, BF₁₀ > 15). In light of this, we proposed a purpose-built quantitative assessment checklist, which demonstrated that the overall ratings of these models increased by publication year but were negatively associated with model performance.

CONCLUSIONS

Together, improving sampling economic equality and hence the quality of machine learning models may be a crucial facet to plausibly translating neuroimaging-based diagnostic classifiers into clinical practice.

Evaluation of Risk of Bias in Neuroimaging-Based Artificial Intelligence Models for Psychiatric Diagnosis: A Systematic Review

IMPORTANCE

Neuroimaging-based artificial intelligence (AI) diagnostic models have proliferated in psychiatry. However, their clinical applicability and reporting quality (ie, feasibility) for clinical practice have not been systematically evaluated.

OBJECTIVE

To systematically assess the risk of bias (ROB) and reporting quality of neuroimaging-based AI models for psychiatric diagnosis.

EVIDENCE REVIEW

PubMed was searched for peer-reviewed, full-length articles published between January 1, 1990, and March 16, 2022. Studies aimed at developing or validating neuroimaging-based AI models for clinical diagnosis of psychiatric disorders were included. Reference lists were further searched for suitable original studies. Data extraction followed the CHARMS (Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies) and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines. A closed-loop cross-sequential design was used for quality control. The PROBAST (Prediction Model Risk of Bias Assessment Tool) and modified CLEAR (Checklist for Evaluation of Image-Based Artificial Intelligence Reports) benchmarks were used to systematically evaluate ROB and reporting quality.

FINDINGS

A total of 517 studies presenting 555 AI models were included and evaluated. Of these models, 461 (83.1%; 95% CI, 80.0%-86.2%) were rated as having a high overall ROB based on the PROBAST. The ROB was particular high in the analysis domain, including inadequate sample size (398 of 555 models [71.7%; 95% CI, 68.0%-75.6%]), poor model performance examination (with 100% of models lacking calibration examination), and lack of handling data complexity (550 of 555 models [99.1%; 95% CI, 98.3%-99.9%]). None of the AI models was perceived to be applicable to clinical practices. Overall reporting completeness (ie, number of reported items/number of total items) for the AI models was 61.2% (95% CI, 60.6%-61.8%), and the completeness was poorest for the technical assessment domain with 39.9% (95% CI, 38.8%-41.1%).

CONCLUSIONS AND RELEVANCE

This systematic review found that the clinical applicability and feasibility of neuroimaging-based AI models for psychiatric diagnosis were challenged by a high ROB and poor reporting quality. Particularly in the analysis domain, ROB in AI diagnostic models should be addressed before clinical application.

Machine Learning Techniques for the Diagnosis of Schizophrenia Based on Event-Related Potentials

Antecedent

The event-related potential (ERP) components P300 and mismatch negativity (MMN) have been linked to cognitive deficits in patients with schizophrenia. The diagnosis of schizophrenia could be improved by applying machine learning procedures to these objective neurophysiological biomarkers. Several studies have attempted to achieve this goal, but no study has examined Multiple Kernel Learning (MKL) classifiers. This algorithm finds optimally a combination of kernel functions, integrating them in a meaningful manner, and thus could improve diagnosis.

Objective

This study aimed to examine the efficacy of the MKL classifier and the Boruta feature selection method for schizophrenia patients (SZ) and healthy controls (HC) single-subject classification.

Methods

A cohort of 54 SZ and 54 HC participants were studied. Three sets of features related to ERP signals were calculated as follows: peak related features, peak to peak related features, and signal related features. The Boruta algorithm was used to evaluate the impact of feature selection on classification performance. An MKL algorithm was applied to address schizophrenia detection.

Results

A classification accuracy of 83% using the whole dataset, and 86% after applying Boruta feature selection was obtained. The variables that contributed most to the classification were mainly related to the latency and amplitude of the auditory P300 paradigm.

Conclusion

This study showed that MKL can be useful in distinguishing between schizophrenic patients and controls when using ERP measures. Moreover, the use of the Boruta algorithm provides an improvement in classification accuracy and computational cost.

A Multimodal Neuroprosthetic Interface to Record, Modulate and Classify Electrophysiological Biomarkers Relevant to Neuropsychiatric Disorders

Most mental disorders, such as addictive diseases or schizophrenia, are characterized by impaired cognitive function and behavior control originating from disturbances within prefrontal neural networks. Their often chronic reoccurring nature and the lack of efficient therapies necessitate the development of new treatment strategies. Brain-computer interfaces, equipped with multiple sensing and stimulation abilities, offer a new toolbox whose suitability for diagnosis and therapy of mental disorders has not yet been explored. This study, therefore, aimed to develop a biocompatible and multimodal neuroprosthesis to measure and modulate prefrontal neurophysiological features of neuropsychiatric symptoms. We used a 3D-printing technology to rapidly prototype customized bioelectronic implants through robot-controlled deposition of soft silicones and a conductive platinum ink. We implanted the device epidurally above the medial prefrontal cortex of rats and obtained auditory event-related brain potentials in treatment-naïve animals, after alcohol administration and following neuromodulation through implant-driven electrical brain stimulation and cortical delivery of the anti-relapse medication naltrexone. Towards smart neuroprosthetic interfaces, we furthermore developed machine learning algorithms to autonomously classify treatment effects within the neural recordings. The neuroprosthesis successfully captured neural activity patterns reflecting intact stimulus processing and alcohol-induced neural depression. Moreover, implant-driven electrical and pharmacological stimulation enabled successful enhancement of neural activity. A machine learning approach based on stepwise linear discriminant analysis was able to deal with sparsity in the data and distinguished treatments with high accuracy. Our work demonstrates the feasibility of multimodal bioelectronic systems to monitor, modulate and identify healthy and affected brain states with potential use in a personalized and optimized therapy of neuropsychiatric disorders.

Classification of Schizophrenia by Combination of Brain Effective and Functional Connectivity

At present, lots of studies have tried to apply machine learning to different electroencephalography (EEG) measures for diagnosing schizophrenia (SZ) patients. However, most EEG measures previously used are either a univariate measure or a single type of brain connectivity, which may not fully capture the abnormal brain changes of SZ patients. In this paper, event-related potentials were collected from 45 SZ patients and 30 healthy controls (HCs) during a learning task, and then a combination of partial directed coherence (PDC) effective and phase lag index (PLI) functional connectivity were used as features to train a support vector machine classifier with leave-one-out cross-validation for classification of SZ from HCs. Our results indicated that an excellent classification performance (accuracy = 95.16%, specificity = 94.44%, and sensitivity = 96.15%) was obtained when the combination of functional and effective connectivity features was used, and the corresponding optimal feature number was 15, which included 12 PDC and three PLI connectivity features. The selected effective connectivity features were mainly located between the frontal/temporal/central and visual/parietal lobes, and the selected functional connectivity features were mainly located between the frontal/temporal and visual cortexes of the right hemisphere. In addition, most of the selected effective connectivity abnormally enhanced in SZ patients compared with HCs, whereas all the selected functional connectivity features decreased in SZ patients. The above results showed that our proposed method has great potential to become a tool for the auxiliary diagnosis of SZ.

Multivariate patterns of EEG microstate parameters and their role in the discrimination of patients with schizophrenia from healthy controls

Quasi-stable electrical fields in the EEG, called microstates carry information on the dynamics of large scale brain networks. Using machine learning techniques, we explored whether abnormalities in microstates can be used to classify patients with schizophrenia and healthy controls. We applied multivariate pattern analysis of microstate features to create a specified feature set to represent microstate characteristics. Machine learning approaches using these features for classification of patients with schizophrenia were compared with prior EEG based machine learning studies. Our microstate segmentation in both patients with schizophrenia and healthy controls yielded topographies that were similar to the normative database established earlier by Koenig et al. Our machine learning model was based on large sample size, low number of features and state-of-art K-fold cross-validation technique. The multivariate analysis revealed three patterns of correlated features, which yielded an AUC of 0.84 for the group separation (accuracy: 82.7%, sensitivity/specificity: 83.5%/85.3%). Microstate segmentation of resting state EEG results in informative features to discriminate patients with schizophrenia from healthy individuals. Moreover, alteration in microstate measures may represent disturbed activity of networks in patients with schizophrenia.

EEG Signals Classification Using Machine Learning for The Identification and Diagnosis of Schizophrenia

This paper presents the design of a machine learning-based classifier for the differentiation between Schizophrenia patients and healthy controls using features extracted from electroencephalograph(EEG) signals based on event related potential(ERP). A number of features are extracted from an online EEG dataset with 81 subjects, including 32 healthy controls and 49 Schizophrenia patients. The EEG signals are preprocessed and since the dataset is relatively small, the random forest machine learning algorithm is chosen to be applied on different combinations of feature sets for classification. It is found that the classification accuracy can be improved by adding certain features extracted from EEG signals.

Applying deep learning to single-trial EEG data provides evidence for complementary theories on action control

Efficient action control is indispensable for goal-directed behaviour. Different theories have stressed the importance of either attention or response selection sub-processes for action control. Yet, it is unclear to what extent these processes can be identified in the dynamics of neurophysiological (EEG) processes at the single-trial level and be used to predict the presence of conflicts in a given moment. Applying deep learning, which was blind to cognitive theory, on single-trial EEG data allowed to predict the presence of conflict in ~95% of subjects ~33% above chance level. Neurophysiological features related to attentional and motor response selection processes in the occipital cortex and the superior frontal gyrus contributed most to prediction accuracy. Importantly, deep learning was able to identify predictive neurophysiological processes in single-trial neural dynamics. Hence, mathematical (artificial intelligence) approaches may be used to foster the validation and development of links between cognitive theory and neurophysiology of human behavior.

Vahid et al. use a deep-learning approach to analyze single-trial EEG data to examine theories on action control. Their approach enables the identification of spatial and temporal neurophysiological features that are predictive of the response control during the Simon task. The results confirm cognitive theory-driven approaches on the relationship between neurophysiology and human behavior.

Cognitive Memory Comparison Between Tinnitus and Normal Cases Using Event-Related Potentials

About 20 percent of people above 60 years old suffer from tinnitus though no definitive treatment has been found for it. Evaluation of electrical brain activity using Event-Related Potentials (ERPs) is one of the methods to investigate the underlying reasons of tinnitus perception. Previous studies using ERPs suggest that the precognitive memory in tinnitus groups is negatively affected in comparison to the normal hearing groups. In this study, cognitive memory has been assessed using visual and auditory P300 response with oddball paradigm. Fifteen chronic tinnitus subjects and six normal hearing subjects participated in the experiment. T-test with significance level of 0.05 was applied on amplitude and latency of auditory and visual P300 for all electroencephalography (EEG) channels separately to compare tinnitus and normal hearing groups where the tinnitus group showed meaningful lower amplitude of auditory P300 peak in three EEG channels.

Binary classification of multichannel-EEG records based on the ϵ-complexity of continuous vector functions

BACKGROUND AND OBJECTIVE

A crucial step in a classification of electroencephalogram (EEG) records is the feature selection. The feature selection problem is difficult because of the complex structure of EEG signals. To classify the EEG signals with good accuracy, most of the recently published studies have used high-dimensional feature spaces. Our objective is to create a low-dimensional feature space that enables binary classification of EEG records.

METHODS

The proposed approach is based on our theory of the ϵ-complexity of continuous functions, which is extended here (see Appendix) to the case of vector functions. This extension permits us to handle multichannel-EEG records. The method consists of two steps. Firstly, we estimate the ϵ-complexity coefficients of the original signal and its finite differences. Secondly, we utilize the random forest (RF) or support vector machine (SVM) classifier.

RESULTS

We demonstrated the performance of our method on simulated data. We also applied it to the problem of classification of multichannel-EEG records related to a group of healthy adolescents (39 subjects) and a group of adolescents with schizophrenia (45 subjects). We found that the random forest classifier provides a superior result. In particular, out-of-bag accuracy in the case of RF was 85.3%. Using 10-fold cross-validation (CV), RF gave an average accuracy of 84.5% on a test set, whereas SVM gave an accuracy of 81.07%. We note that the highest accuracy on CV was 89.3%. To compare our method with the classical approach, we performed classification using the spectral features. In this case, the best performance was achieved using seven-dimensional feature space, with an average accuracy of 83.6%.

CONCLUSIONS

We developed a model-free method for binary classification of EEG records. The feature space was reduced to four dimensions. The results obtained indicate the effectiveness of the proposed method.

Machine-learning-based diagnosis of schizophrenia using combined sensor-level and source-level EEG features

Recently, an increasing number of researchers have endeavored to develop practical tools for diagnosing patients with schizophrenia using machine learning techniques applied to EEG biomarkers. Although a number of studies showed that source-level EEG features can potentially be applied to the differential diagnosis of schizophrenia, most studies have used only sensor-level EEG features such as ERP peak amplitude and power spectrum for machine learning-based diagnosis of schizophrenia. In this study, we used both sensor-level and source-level features extracted from EEG signals recorded during an auditory oddball task for the classification of patients with schizophrenia and healthy controls. EEG signals were recorded from 34 patients with schizophrenia and 34 healthy controls while each subject was asked to attend to oddball tones. Our results demonstrated higher classification accuracy when source-level features were used together with sensor-level features, compared to when only sensor-level features were used. In addition, the selected sensor-level features were mostly found in the frontal area, and the selected source-level features were mostly extracted from the temporal area, which coincide well with the well-known pathological region of cognitive processing in patients with schizophrenia. Our results suggest that our approach would be a promising tool for the computer-aided diagnosis of schizophrenia.

Single-subject prediction of response inhibition behavior by event-related potentials

Much research has been devoted to investigating response inhibition and the neuronal processes constituting this essential cognitive faculty. However, the nexus between cognitive subprocesses, behavior, and electrophysiological processes remains associative in nature. We therefore investigated whether neurophysiological correlates of inhibition subprocesses merely correlate with behavioral performance or actually provide information expedient to the prediction of behavior on a single-subject level. Tackling this question, we used different data-driven classification approaches in a sample of n = 262 healthy young subjects who completed a standard Go/Nogo task while an EEG was recorded. On the basis of median-split response inhibition performance, subjects were classified as "accurate/slow" and "less accurate/fast." Even though these behavioral group differences were associated with significant amplitude variations in classical electrophysiological correlates of response inhibition (i.e., N2 and P3), they were not predictive for group membership on a single-subject level. Instead, amplitude differences in the Go-P2 originating in the precuneus (BA7) were shown to predict group membership on a single-subject level with up to 64% accuracy. These findings strongly suggest that the behavioral outcome of response inhibition greatly depends on the amount of cognitive resources allocated to early stages of stimulus-response activation during responding. This suggests that research should focus more on early processing steps during responding when trying to understand the origin of interindividual differences in response inhibition processes.

Progressive reduction of visual P300 amplitude in patients with first-episode schizophrenia: an ERP study

OBJECTIVE

To understand the underlying dynamic neurophysiological changes over the course of schizophrenia, it is important to study subjects longitudinally from the early stage of the illness. We previously reported that visual P300 was already impaired in patients with first-episode schizophrenia (FESZ). This study demonstrates how the visual P300, as well as earlier components P1, N1, and N200, changed at the 1-year follow-up after their initial measurement.

METHODS

Visual ERPs were recorded with the same experimental paradigm and acquisition protocol at both time points in FESZ (n = 18) and healthy comparison subjects (n = 24). Participants silently counted infrequent target stimuli ("x") amid standard stimuli ("y") presented on the screen while the 64-channel electroencephalogram was recorded.

RESULTS

FESZ showed smaller visual P300, N200, P1 (trend level) amplitude and delayed P300 and N1 latency at both time points; however, only P300 showed progressive amplitude reduction over the course of the illness at 1-year follow-up. P300 latency did not change over time in either group. FESZ showed significantly reduced Spatial Span total score at both time points, and there was a significant negative correlation between P300 peak amplitude and the Brief Psychiatric Rating Scale positive symptom score at baseline.

CONCLUSION

These data show progressive P300 amplitude reduction in response to visual stimuli in the early stage of schizophrenia. These visual P300 findings support the concept of progression of schizophrenia, suggesting the usefulness of the visual P300 as a biological marker of progression.

Comparison of auditory and visual oddball fMRI in schizophrenia

Individuals with schizophrenia often suffer from attentional deficits, both in focusing on task-relevant targets and in inhibiting responses to distractors. Schizophrenia also has a differential impact on attention depending on modality: auditory or visual. However, it remains unclear how abnormal activation of attentional circuitry differs between auditory and visual modalities, as these two modalities have not been directly compared in the same individuals with schizophrenia. We utilized event-related functional magnetic resonance imaging (fMRI) to compare patterns of brain activation during an auditory and visual oddball task in order to identify modality-specific attentional impairment. Healthy controls (n=22) and patients with schizophrenia (n=20) completed auditory and visual oddball tasks in separate sessions. For responses to targets, the auditory modality yielded greater activation than the visual modality (A-V) in auditory cortex, insula, and parietal operculum, but visual activation was greater than auditory (V-A) in visual cortex. For responses to novels, A-V differences were found in auditory cortex, insula, and supramarginal gyrus; and V-A differences in the visual cortex, inferior temporal gyrus, and superior parietal lobule. Group differences in modality-specific activation were found only for novel stimuli; controls showed larger A-V differences than patients in prefrontal cortex and the putamen. Furthermore, for patients, greater severity of negative symptoms was associated with greater divergence of A-V novel activation in the visual cortex. Our results demonstrate that patients have more pronounced activation abnormalities in auditory compared to visual attention, and link modality specific abnormalities to negative symptom severity.

Critical evaluation of auditory event-related potential deficits in schizophrenia: evidence from large-scale single-subject pattern classification

Event-related potential (ERP) deficits associated with auditory oddball and click-conditioning paradigms are among the most consistent findings in schizophrenia and are discussed as potential biomarkers. However, it is unclear to what extend these ERP deficits distinguish between schizophrenia patients and healthy controls on a single-subject level, which is of high importance for potential translation to clinical routine. Here, we investigated 144 schizophrenia patients and 144 matched controls with an auditory click-conditioning/oddball paradigm. P50 and N1 gating ratios as well as target-locked N1 and P3 components were submitted to conventional general linear models and to explorative machine learning algorithms. Repeated-measures ANOVAs revealed significant between-group differences for the oddball-locked N1 and P3 components but not for any gating measure. Machine learning-assisted analysis achieved 77.7% balanced classification accuracy using a combination of target-locked N1 and P3 amplitudes as classifiers. The superiority of machine learning over repeated-measures analysis for classifying schizophrenia patients was in the range of about 10% as quantified by receiver operating characteristics. For the first time, our study provides large-scale single-subject classification data on auditory click-conditioning and oddball paradigms in schizophrenia. Although our study exemplifies how automated inference may substantially improve classification accuracy, our data also show that the investigated ERP measures show comparably poor discriminatory properties in single subjects, thus illustrating the need to establish either new analytical approaches for these paradigms or other paradigms to investigate the disorder.

Neurocognitive pattern analysis reveals classificatory hierarchy of attention deficits in schizophrenia

Attention deficits, among other cognitive deficits, are frequently observed in schizophrenia. Although valid and reliable neurocognitive tasks have been established to assess attention deficits in schizophrenia, the hierarchical value of those tests as diagnostic discriminants on a single-subject level remains unclear. Thus, much research is devoted to attention deficits that are unlikely to be translated into clinical practice. On the other hand, a clear hierarchy of attention deficits in schizophrenia could considerably aid diagnostic decisions and may prove beneficial for longitudinal monitoring of therapeutic advances. To propose a diagnostic hierarchy of attention deficits in schizophrenia, we investigated several facets of attention in 86 schizophrenia patients and 86 healthy controls using a set of established attention tests. We applied state-of-the-art machine learning algorithms to determine attentive test variables that enable an automated differentiation between schizophrenia patients and healthy controls. After feature preranking, hypothesis building, and hypothesis validation, the polynomial support vector machine classifier achieved a classification accuracy of 90.70% ± 2.9% using psychomotor speed and 3 different attention parameters derived from sustained and divided attention tasks. Our study proposes, to the best of our knowledge, the first hierarchy of attention deficits in schizophrenia by identifying the most discriminative attention parameters among a variety of attention deficits found in schizophrenia patients. Our results offer a starting point for hierarchy building of schizophrenia-associated attention deficits and contribute to translating these concepts into diagnostic and therapeutic practice on a single-subject level.