Hyperdimensional Computing for ADHD Classification using EEG Signals

Federica Colonnese, Antonello Rosato, Francesco Di Luzio, Massimo Panella · 2025

Read Hyperdimensional Computing for ADHD Classification using EEG Signals by Federica Colonnese, Antonello Rosato, Francesco Di Luzio, Massimo Panella (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-05186.txt.

Introduction

Federica Colonnese, Antonello Rosato, Francesco Di Luzio, and Massimo Panella present a 2025 study on detecting ADHD from brain wave recordings. They test hyperdimensional computing, a method that encodes data as very long vectors and learns with simple operations. Their model reads the time and space patterns of EEG signals, the traces of brain activity taken from the scalp. No earlier work had applied this approach to ADHD classification.

The model reached accuracy near eighty-nine percent, beating standard deep neural network rivals on the same task. More striking was its thrift with data: trained on just seven of seventy-nine patients, it matched rivals trained on everyone. That efficiency matters in clinics, where large labeled data sets are rare. The authors present the method as a fast, light route to computer-aided ADHD diagnosis.

Catalog record

CREATORFederica Colonnese, Antonello Rosato, Francesco Di Luzio, Massimo Panella
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-05186

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Hyperdimensional Computing for ADHD Classification using EEG Signals Federica Colonnese, Antonello Rosato, Francesco Di Luzio, Massimo Panella Abstract Following the recent interest in applying the Hyperdimensional Computing paradigm in medical context to power up the performance of general machine learning applied to biomedical data, this study represents the first attempt at employing such techniques to solve the problem of classification of Attention Deficit Hyperactivity Disorder using electroencephalogram signals. Making use of a spatio-temporal encoder, and leveraging the properties of HDC, the proposed model achieves an accuracy of 88.9 % 88.9\% , outperforming traditional Deep Neural Networks benchmark models. The core of this research is not only to enhance the classification accuracy of the model but also to explore its efficiency in terms of the required training data: a critical finding of the study is the identification of the minimum number of patients needed in the training set to achieve a sufficient level of accuracy. To this end, the accuracy of our model trained with only 7 7 of the 79 79 patients is comparable to the one from benchmarks trained on the full dataset. This finding underscores the model’s efficiency and its potential for quick and precise ADHD diagnosis in medical settings where large datasets are typically unattainable. I Introduction Hyperdimensional Computing (HDC) techniques, also called Vector Symbolic Architectures (VSA), have gained increasing attention in the fields of Artificial Intelligence (AI) and neural sciences. This brain-inspired computational approach has been widely applied in a growing number of domains, including signal recognition [ 1 ] , biomedical and image classification [ 2 , 3 ] . It draws inspiration from neuroscience research [ 4 ] , and has showcased significant promises in tackling intricate challenges related to data representation and processing, which nowadays are areas largely ruled by Deep Neural Networks (DNNs). DNNs encounter significant challenges as their computational expenses can be heavy, especially for complex problems: the exponential increase in parameters leads to higher computational complexity and memory requirements, prolonged training times due to the optimization of numerous parameters, and the need for large datasets. The scientific community is exploring integrating HDC principles with DNN frameworks to address, for example, complexity and data requirements. HDC involves projecting data onto thousands or millions of dimensions, moving away from traditional numerical representations, and fundamentally changing the way information is handled [ 4 ] . As a result, HDC techniques are increasingly employed in a multitude of domains, ranging from the biomedical field [ 5 ] , to the identification of driving styles [ 6 ] , as well as speech recognition [ 1 ] , highlighting the adaptability and significance of HDC methodologies across various sectors. HDC is attracting interest for its potential in scenarios needing alternatives to current techniques, particularly due to its ability to handle complex data situations effectively. This paper examines an application of HDC to the biomedical domain, specifically to the detection of Attention Deficit Hyperactivity Disorder (ADHD) using Electroencephalogram (EEG) signals, recorded while patients were resting with open eyes. To this end, ADHD, one of the most prevalent neurodevelopmental disorders, impacts 4\%-12\% of young children and 4\%-5\% of adults [ 7 ] . According to the criteria established by the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) [ 8 ] , diagnosing ADHD generally depends on subjective methods. These include behavioral assessments to evaluate how individuals socialize, behave, and communicate, along with standardized diagnostic tools such as questionnaires completed by parents and teachers to observe behavior across different settings. However, ADHD’s misdiagnosis is common due to symptoms overlapping with other conditions and neurodevelopmental disorders itself, insufficient training among healthcare professionals, and the stigma associated with this condition leading also to under-reporting, posing challenges for accurate diagnosis procedures [ 9 ] . To the best of our knowledge, this study represents the first attempt at integrating HDC with ADHD classification. On this end, the key distinction and main novelty aspects lie, not only in achieving higher accuracy rates compared to DNNs but, mostly in accomplishing this with a reduced amount of patient data inputs: by training the proposed framework using EEG signals recorded on a very small subset of patients (seven patients, meaning the 8.86\% of the dataset), it was possible to obtain a classification accuracy of 72.9\% , already surpassing the benchmark DNNs trained on the whole dataset. These improvements led to a quick, efficient tool that could merge DSM-5 methods with automation, speeding up diagnoses, enhancing accuracy, and reducing the workload for healthcare professionals by simplifying processes and decreasing data preparation. The code developed in this research will be made available upon request. II Current Literature and Main Goals This section briefly reviews current HDC applications for neurodevelopmental disorders and introduces the novel contributions proposed by our approach in this field. II-A State of The Art In the biomedical domain, HDC classification methods have demonstrated exceptional performances across different applications, including but not limited to seizure detection [ 10 ] , gesture recognition [ 11 ] , EEG error-related potentials classification for brain-computer interfaces [ 12 ] , and emotion recognition [ 13 ] . Furthermore, it has exhibited remarkable classification accuracy even when confronted with challenges such as limited training data [ 14 ] , one-shot learning situations [ 15 ] , and frequent model updates. These characteristics present significant advantages for medical physiological sensors tailored to user-specific data, given the susceptibility of these sensors to displacements between runs or during extended periods of use [ 16 ] . The growing interest in biomedical research for ADHD detection is largely driven by advancements in medical technology. Imaging techniques are frequently employed in this field and DNNs are the primary methods for analyzing these complex datasets due to their ability to identify intricate patterns as, for instance, in brain activity from functional magnetic resonance imaging (fMRI) [ 17 , 18 , 19 , 20 ] and EEG [ 21 , 22 ] , as well as in extended behavioral analysis and emotion recognition from multivariate data and time series [ 23 , 24 ] . However, DNNs encounter challenges such as limited dataset availability and the need for complex, resource-intensive networks to effectively process those large-scale medical data. Despite these advancements, to the best of our knowledge, there is currently no known application of HDC for the classification of ADHD or neurodevelopmental disorders in general. As will be shown in the following paragraphs, HDC could offer potential benefits in this area due to its efficiency and robustness in handling high-dimensional data. Exploring HDC might lead to new, less resource-intensive methods for diagnosing and understanding ADHD, opening new opportunities for research and application in neurodevelopmental studies. Recent advancements in the classification of ADHD, have leveraged the potential of Transfer Learning (TL) and pre-trained Convolutional Neural Network (CNN) models [ 25 ] . These approaches offer a robust framework for overcoming challenges associated with high-dimensional data, variances in data acquisition, and class distribution imbalances. Specifically, Transfer Learning has shown significant promise in scenarios with limited labeled data by generalizing patterns learned from large datasets to task-specific applications. For instance, [ 26 ] demonstrated the effectiveness of TL using the ResNet-50 architecture to classify ADHD from functional Magnetic

Read the complete text