TY - GEN
T1 - Privacy-Preserving EEG Signal Analysis with Electrode Attention for Depression Diagnosis
T2 - 2023 IEEE Global Communications Conference, GLOBECOM 2023
AU - Dong, Huanze
AU - Wu, Jun
AU - Bashir, Ali Kashif
AU - Pan, Qianqian
AU - Omar, Marwan
AU - Al-Dulaimi, Anwer
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Artificial intelligence has been utilized to analyze patients' electroencephalograms (EEG) to diagnose depression. However, attackers can deduce patients' privacy after analyzing patients' EEG time series. Therefore, researchers propose to operate ciphertext calculation in depression diagnosis models based on homomorphic encryption. Nevertheless, homomorphic encryption requires consistent private keys during training, which could result in other participants decrypting the ci-phertexts. Additionally, existing EEG-based models neglect the relationship among electrode positions during EEG acquisition. To address these issues, we propose a novel training strategy for the depression diagnosis model based on fully homomorphic en-cryption (FHE) and electrode topology. Specifically, we establish a training strategy that prioritizes the privacy of patients' EEG data without compromising the cost-effectiveness of the diagnosis model. Furthermore, we incorporate the attention mechanism of electrode topology into our model to improve its performance and verify the relationship among topology locations. Our proposed model outperforms the original convolution neural network model, achieving higher accuracy in depression diagnosis and identifying virtual electrode channels for the first time.
AB - Artificial intelligence has been utilized to analyze patients' electroencephalograms (EEG) to diagnose depression. However, attackers can deduce patients' privacy after analyzing patients' EEG time series. Therefore, researchers propose to operate ciphertext calculation in depression diagnosis models based on homomorphic encryption. Nevertheless, homomorphic encryption requires consistent private keys during training, which could result in other participants decrypting the ci-phertexts. Additionally, existing EEG-based models neglect the relationship among electrode positions during EEG acquisition. To address these issues, we propose a novel training strategy for the depression diagnosis model based on fully homomorphic en-cryption (FHE) and electrode topology. Specifically, we establish a training strategy that prioritizes the privacy of patients' EEG data without compromising the cost-effectiveness of the diagnosis model. Furthermore, we incorporate the attention mechanism of electrode topology into our model to improve its performance and verify the relationship among topology locations. Our proposed model outperforms the original convolution neural network model, achieving higher accuracy in depression diagnosis and identifying virtual electrode channels for the first time.
KW - Artificial Intelligence
KW - Depression Diagnosis
KW - Electroencephalogram
KW - Fully Homomorphic En-cryption
KW - Privacy Preservation
UR - http://www.scopus.com/inward/record.url?scp=85187379770&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85187379770&partnerID=8YFLogxK
U2 - 10.1109/GLOBECOM54140.2023.10436783
DO - 10.1109/GLOBECOM54140.2023.10436783
M3 - Conference contribution
AN - SCOPUS:85187379770
T3 - Proceedings - IEEE Global Communications Conference, GLOBECOM
SP - 4265
EP - 4270
BT - GLOBECOM 2023 - 2023 IEEE Global Communications Conference
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 4 December 2023 through 8 December 2023
ER -