Generative Perturbation Network for Universal Adversarial Attacks on Brain-Computer Interfaces

Jung, Jongwan; P., Moon, H.; Yu, Geunhyeok; Hwang, Hyoseok

doi:10.1109/jbhi.2023.3303494

Cited by 5 publications

(1 citation statement)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Tsai et al [2] conducted a one-pixel attack on various medical image datasets, such as COVID-19, Chest, derma and pneumonia to generate adversarial images that can fool a trained model. It also performed a multi-pixel attack on the COVID-19 dataset to explore the impact of the number of perturbed pixels.…”

Section: Literature Surveymentioning

confidence: 99%

Adversarial Attacks on Medical Image Diagnosis Models And its Mitigation Techniques

Sahu,

Prasanna,

Neelima

et al. 2024

Int. J. Res. Publ. Rev.

View full text Add to dashboard Cite

In recent years, deep learning (DL) models have become integral to numerous sectors, revolutionizing our daily lives and workflows. Particularly in healthcare, DL have brought about a paradigm shift in medical diagnosis through innovative image analysis capabilities. These computational tools offer exceptional precision and speed, significantly enhancing diagnostic accuracy and facilitating early disease detection. However, the widespread adoption and reliance on these models have opened the door to new forms of vulnerabilities, notably adversarial attacks. In the context of medical image diagnosis, adversarial attacks pose an alarming threat. They can manipulate diagnostic models into misinterpreting imaging data, leading to false positives or negatives. Such errors can result in misdiagnosis, delayed treatment, or unnecessary interventions, impacting patient safety and the overall quality of healthcare. This project delves into the landscape of adversarial attacks in the context of medical image diagnosis. This study looks at popular adversarial attack strategies. A Convolutional Neural Network (CNN) EfficientNet B0 Model trained to categorise Alzheimer's brain MRI images is subjected to the Vertical Perturbation attack, Fast Gradient sign Method (FGSM), and Square attack.Following that, it goes into one of the most common adversarial defence approaches, adversarial training. The performance of the model that has been trained on adversarial instances is next tested against the previously described attacks, and recommendations to improve the neural network's robustness are therefore supplied based on the experiment findings.

show abstract

Section: Literature Surveymentioning

confidence: 99%

Adversarial Attacks on Medical Image Diagnosis Models And its Mitigation Techniques

Sahu,

Prasanna,

Neelima

et al. 2024

Int. J. Res. Publ. Rev.

View full text Add to dashboard Cite

show abstract

Spatialspectral-Backdoor: Realizing backdoor attack for deep neural networks in brain–computer interface via EEG characteristics

Li,

Huang,

You

et al. 2025

Neurocomputing

View full text Add to dashboard Cite

Adversarial artifact detection in EEG-based brain–computer interfaces

Chen,

Meng,

et al. 2024

J. Neural Eng.

View full text Add to dashboard Cite

Objective: Machine learning has achieved significant success in electroencephalogram (EEG) based brain-computer interfaces (BCIs), with most existing research focusing on improving the decoding accuracy. However, recent studies have shown that EEG-based BCIs are vulnerable to adversarial attacks, where small perturbations added to the input can cause misclassification. Detecting adversarial examples is crucial for both understanding this phenomenon and developing effective defense strategies. Approach: This paper, for the first time, explores adversarial detection in EEG-based BCIs. We extend several popular adversarial detection approaches from computer vision to BCIs. Two new Mahalanobis distance based adversarial detection approaches, and three cosine distance based adversarial detection approaches, are also proposed, which showed promising performance in detecting three kinds of white-box attacks. Main results: We evaluated the performance of eight adversarial detection approaches on three EEG datasets, three neural networks, and four types of adversarial attacks. Our approach achieved an area under the curve (AUC) score of up to 99.99% in detecting white-box attacks. Additionally, we assessed the transferability of different adversarial detectors to unknown attacks. Significance: Through extensive experiments, we found that white-box attacks may be easily detected, and differences exist in the distributions of different types of adversarial examples. Our work should facilitate understanding the vulnerability of existing BCI models and developing more secure BCIs in the future.

show abstract

Generative Perturbation Network for Universal Adversarial Attacks on Brain-Computer Interfaces

Cited by 5 publications

References 39 publications

Adversarial Attacks on Medical Image Diagnosis Models And its Mitigation Techniques

Adversarial Attacks on Medical Image Diagnosis Models And its Mitigation Techniques

Spatialspectral-Backdoor: Realizing backdoor attack for deep neural networks in brain–computer interface via EEG characteristics

Adversarial artifact detection in EEG-based brain–computer interfaces

Contact Info

Product

Resources

About