The analysis of electroencephalographic (EEG) signals has been a fundamental tool for decades in diagnosing neurological disorders, from epilepsy to cognitive decline. However, traditional methods have significant limitations: the use of predefined spectral features introduces a strong bias by determining in advance which patterns are relevant, while deep networks and foundation models require massive amounts of data and computational power, and are difficult for clinicians to interpret. In this context, the bag-of-waves approach emerges as a disruptive alternative that combines interpretability, efficiency, and competitive performance even in very limited data scenarios.
Bag-of-waves proposes building a small dictionary of recurring EEG waveform templates — called atoms — using a shift-invariant k-means algorithm, without labels. Once these atoms are learned, the continuous signal is converted into a sequence of atom tokens, and the frequencies of these tokens feed a simple classifier or clustering step. This representation is enriched by two key extensions: n-grams of transitions between atoms, which capture temporal structure, and multichannel spatial atoms that integrate regional and cross-channel information across electrodes. The result is a model that not only matches the performance of deep networks and foundation models on demanding benchmarks like TUEV (six-way classification of clinical events), but does so with a fraction of the parameters and offers full interpretability: each atom corresponds to an inspectable waveform morphology, directly verifiable by neurophysiologists.
The great advantage of bag-of-waves lies in its ability to operate in low-data regimes, where heavy models fail due to lack of samples. For example, in a study with only sixteen mice (genotype clustering), the method achieved competitive results, demonstrating its suitability for clinical applications where collecting large volumes of labeled EEG is infeasible or extremely costly. Moreover, its modular architecture allows adaptation to different needs: from resting-state dementia classification to epileptic event detection, all with a waveform dictionary that can be analyzed and refined by experts.
From a technical and business perspective, bag-of-waves represents a paradigm shift in the development of custom software for the healthcare sector. Combining this methodology with cloud platforms (AWS, Azure) allows scaling EEG signal processing without compromising privacy or latency, while integration with Business Intelligence tools like Power BI facilitates pattern visualization and automated clinical reporting. Cybersecurity, meanwhile, is an unavoidable pillar when handling sensitive patient data, and implementing robust encryption and access control protocols is critical here.
At Q2BSTUDIO, a company specialized in software and technology development, we understand that innovation in AI-assisted diagnosis requires a pragmatic approach. Our team can build bag-of-waves-based systems that integrate with hospital workflows, using cloud infrastructures for dictionary training and AI agents that automate anomaly detection. Furthermore, the interpretable nature of the method facilitates clinical validation and regulatory certification, key aspects for real adoption in medical environments.
The results of bag-of-waves open the door to new artificial intelligence services applied to biosignal analysis, with a focus on transparency and computational efficiency. Whether for classifying sleep stages, detecting epileptic seizures, or monitoring the progression of neurodegenerative diseases, this technique offers a clear path toward clinical tools that doctors can understand and trust. The combination of waveform dictionaries, temporal n-grams, and spatial atoms turns EEG into a structured language, where each pattern has a verifiable physiological meaning.
In short, bag-of-waves demonstrates that it is not always necessary to resort to massive models to achieve state-of-the-art results. With a lightweight, interpretable architecture trainable on few data, it positions itself as an ideal solution for startups and research centers looking to develop EEG biomarkers in an agile and reliable way. At Q2BSTUDIO, we offer the technical capabilities to transform this methodology into robust software products, integrating cloud, cybersecurity, and data analytics, all with the goal of accelerating the adoption of artificial intelligence in neurology.
