Our Science

The first EEG-specific quantum machine learning paper demonstrated that quantum machine learning can, to some extent, enhance the encoding capability of noisy temporal information.

The abstract introduces Quantum-EEGNet (QEEGNet), a novel hybrid neural network that integrates quantum computing with the traditional EEGNet architecture to improve the encoding and analysis of EEG signals. While acknowledging that QEEGNet's results might not always surpass traditional methods, the paper demonstrates its potential by showing that QEEGNet consistently outperforms EEGNet on most subjects and exhibits greater robustness to noise when evaluated on a benchmark EEG dataset (BCI Competition IV 2a). The study highlights the significant potential of quantum-enhanced neural networks in EEG analysis, pointing to new research and practical applications in the field.

Paper Link

This interesting study attempts to enable us to make significant technical advancements in EEG-image understanding!

This paper introduces the MUSE (MUltimodal Similarity-keeping contrastivE learning) framework, designed to address challenges in decoding images from non-invasive EEG signals, particularly issues related to signal-to-noise ratio and nonstationarity. The framework is used for zero-shot EEG-based image classification, utilizing a series of multivariate time-series encoders specifically tailored for EEG signals. Through regularized contrastive EEG-Image pretraining on an extensive visual EEG dataset, the method achieves state-of-the-art performance, with a top-1 accuracy of 19.3% and a top-5 accuracy of 48.8% in a 200-way zero-shot image classification task. Additionally, the paper includes visualizations of neural patterns via model interpretation, offering insights into the brain's visual processing dynamics.

Paper Link

The first high-impact medical journal (Journal of Affective Disorders, JAD) study demonstrated that machine learning models, using both linear and non-linear EEG features, can effectively predict patients' response rates to different TMS parameters!

The study comparing the prediction accuracy of treatment outcomes for treatment-resistant depression (TRD) using repetitive transcranial magnetic stimulation (rTMS) and intermittent theta-burst stimulation (iTBS). The study's primary goal was to compare the prediction accuracy between frontal theta activity alone and machine learning (ML) models using linear and non-linear frontal EEG signals. Two datasets, one from a randomized controlled trial and another from outpatient data, were analyzed using various ML algorithms, including support vector machine (SVM), random forest (RF), XGBoost, and CatBoost. The results showed that combining EEG features improved prediction accuracy, with the RF model providing the highest accuracy for both rTMS and iTBS. Additionally, patients who received pretreatment predictions had significantly better response rates than those without. The study concludes that ML models, particularly RF, can effectively predict treatment outcomes for TRD using combined EEG features.

Paper Link

The beginning of our EEGxAI algorithmic technology journey originated from the master's thesis methodology of our team's CTO!

The study discusses research on predicting the antidepressant response of patients with Major Depressive Disorder (MDD) to transcranial magnetic stimulation (TMS) using machine learning, deep learning and quantum machine learning models. The study focuses on two types of TMS: repetitive TMS and intermittent theta-burst stimulation. The research utilizes EEG data from 129 MDD patients to train both traditional and quantum machine learning models. A new deep learning model, ACTSNet, incorporating an attention mechanism, showed superior predictive performance compared to other models. Additionally, a proposed data pre-processing technique, booster transformation, further enhances the model's sensitivity. The work also concludes that quantum machine learning on a real quantum computer outperformed traditional algorithms in predicting treatment efficacy.