Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models

Manfred Hartmann; Lukas Lechner; Asbjoern Wulff Helge; Esben Ahrens; Martin Bachler; Bernhard Hametner; Gerhard Gritsch; Tilmann Kluge; Manfred Hartmann

doi:10.1109/bsn58485.2023.10331106

Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models

Manfred Hartmann (Speaker), Lukas Lechner, Asbjoern Wulff Helge, Esben Ahrens, Martin Bachler, Bernhard Hametner, Gerhard Gritsch, Tilmann Kluge, Manfred Hartmann

Medical Signal Analysis

Research output: Chapter in Book or Conference Proceedings › Conference Proceedings with Poster Presentation › peer-review

Abstract

The increasing availability of minimally invasive electroencephalogram (EEG) devices for ultra-long-term recordings has opened new possibilities for advanced EEG analysis, but the resulting large amount of generated data leads to a strong need for computational analyses. Deep neural networks (DNNs) have shown to be powerful for this purpose, but the lack of annotated data from these novel devices is a barrier to DNN training. We propose a novel technique based on fine-tuning of linear pre-processing filters, which is capable of compensating for variations in electrode positions and amplifier characteristics and enables training of models for subcutaneous EEG on largely available scalp EEG data. The effectiveness of the method is demonstrated on a state-of-the-art EEG-based sleep scoring model, where we show that the performance on a database used for training can be retained on the subcutaneous EEG by fine-tuning on data from only three subjects.

Original language	English
Title of host publication	2023 IEEE 19th International Conference on Body Sensor Networks (BSN)
Subtitle of host publication	Conference Proceedings
Place of Publication	Boston, Massachusetts, USA
Chapter	Monday, October 9, 2023
Pages	1-4
Number of pages	4
DOIs	https://doi.org/10.1109/bsn58485.2023.10331106
Publication status	Published - 9 Oct 2023
Event	IEEE-EMBS International Conference on Body Sensor Networks – Sensor and Systems for Digital Health: Sensors and Systems for Digital Health. - MIT Media Lab, Boston, Cambridge, United States Duration: 9 Oct 2023 → 11 Oct 2023 https://bsn.embs.org/2023/

Publication series

Name	International Workshop on Wearable and Implantable Body Sensor Networks (BSN)
Publisher	IEEE
Volume	15

Conference

Conference	IEEE-EMBS International Conference on Body Sensor Networks – Sensor and Systems for Digital Health
Abbreviated title	IEEE BSN 2023
Country/Territory	United States
City	Boston, Cambridge
Period	9/10/23 → 11/10/23
Internet address	https://bsn.embs.org/2023/

Research Field

Medical Signal Analysis

Keywords

deep learning
eeg
wearable devices
sleep scoring

Access to Document

10.1109/bsn58485.2023.10331106

Cite this

Hartmann, M., Lechner, L., Helge, A. W., Ahrens, E., Bachler, M., Hametner, B., Gritsch, G., Kluge, T., & Hartmann, M. (2023). Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models. In 2023 IEEE 19th International Conference on Body Sensor Networks (BSN): Conference Proceedings (pp. 1-4). Article [P48] (International Workshop on Wearable and Implantable Body Sensor Networks (BSN); Vol. 15).. https://doi.org/10.1109/bsn58485.2023.10331106

Hartmann, Manfred ; Lechner, Lukas ; Helge, Asbjoern Wulff et al. / Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models. 2023 IEEE 19th International Conference on Body Sensor Networks (BSN): Conference Proceedings. Boston, Massachusetts, USA, 2023. pp. 1-4 (International Workshop on Wearable and Implantable Body Sensor Networks (BSN)).

@inbook{44912c562879487cafbbfa41f9bb5405,

title = "Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models",

abstract = "The increasing availability of minimally invasive electroencephalogram (EEG) devices for ultra-long-term recordings has opened new possibilities for advanced EEG analysis, but the resulting large amount of generated data leads to a strong need for computational analyses. Deep neural networks (DNNs) have shown to be powerful for this purpose, but the lack of annotated data from these novel devices is a barrier to DNN training. We propose a novel technique based on fine-tuning of linear pre-processing filters, which is capable of compensating for variations in electrode positions and amplifier characteristics and enables training of models for subcutaneous EEG on largely available scalp EEG data. The effectiveness of the method is demonstrated on a state-of-the-art EEG-based sleep scoring model, where we show that the performance on a database used for training can be retained on the subcutaneous EEG by fine-tuning on data from only three subjects.",

keywords = "deep learning, eeg, wearable devices, sleep scoring",

author = "Manfred Hartmann and Lukas Lechner and Helge, {Asbjoern Wulff} and Esben Ahrens and Martin Bachler and Bernhard Hametner and Gerhard Gritsch and Tilmann Kluge and Manfred Hartmann",

note = "{\textcopyright} 2023 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE; IEEE-EMBS International Conference on Body Sensor Networks – Sensor and Systems for Digital Health : Sensors and Systems for Digital Health., IEEE BSN 2023 ; Conference date: 09-10-2023 Through 11-10-2023",

year = "2023",

month = oct,

day = "9",

doi = "10.1109/bsn58485.2023.10331106",

language = "English",

isbn = "979-8-3503-3841-6",

series = "International Workshop on Wearable and Implantable Body Sensor Networks (BSN)",

publisher = "IEEE",

pages = "1--4",

booktitle = "2023 IEEE 19th International Conference on Body Sensor Networks (BSN)",

url = "https://bsn.embs.org/2023/",

}

Hartmann, M , Lechner, L, Helge, AW, Ahrens, E, Bachler, M , Hametner, B , Gritsch, G , Kluge, T & Hartmann, M 2023, Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models. in 2023 IEEE 19th International Conference on Body Sensor Networks (BSN): Conference Proceedings., [P48], International Workshop on Wearable and Implantable Body Sensor Networks (BSN), vol. 15, Boston, Massachusetts, USA, pp. 1-4, IEEE-EMBS International Conference on Body Sensor Networks – Sensor and Systems for Digital Health, Boston, Cambridge, Massachusetts, United States, 9/10/23. https://doi.org/10.1109/bsn58485.2023.10331106

Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models. / Hartmann, Manfred (Speaker); Lechner, Lukas; Helge, Asbjoern Wulff et al.
2023 IEEE 19th International Conference on Body Sensor Networks (BSN): Conference Proceedings. Boston, Massachusetts, USA, 2023. p. 1-4 [P48] (International Workshop on Wearable and Implantable Body Sensor Networks (BSN); Vol. 15).

Research output: Chapter in Book or Conference Proceedings › Conference Proceedings with Poster Presentation › peer-review

TY - CHAP

T1 - Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models

AU - Lechner, Lukas

AU - Helge, Asbjoern Wulff

AU - Ahrens, Esben

AU - Bachler, Martin

AU - Hametner, Bernhard

AU - Gritsch, Gerhard

AU - Kluge, Tilmann

AU - Hartmann, Manfred

A2 - Hartmann, Manfred

N1 - © 2023 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE

PY - 2023/10/9

Y1 - 2023/10/9

N2 - The increasing availability of minimally invasive electroencephalogram (EEG) devices for ultra-long-term recordings has opened new possibilities for advanced EEG analysis, but the resulting large amount of generated data leads to a strong need for computational analyses. Deep neural networks (DNNs) have shown to be powerful for this purpose, but the lack of annotated data from these novel devices is a barrier to DNN training. We propose a novel technique based on fine-tuning of linear pre-processing filters, which is capable of compensating for variations in electrode positions and amplifier characteristics and enables training of models for subcutaneous EEG on largely available scalp EEG data. The effectiveness of the method is demonstrated on a state-of-the-art EEG-based sleep scoring model, where we show that the performance on a database used for training can be retained on the subcutaneous EEG by fine-tuning on data from only three subjects.

AB - The increasing availability of minimally invasive electroencephalogram (EEG) devices for ultra-long-term recordings has opened new possibilities for advanced EEG analysis, but the resulting large amount of generated data leads to a strong need for computational analyses. Deep neural networks (DNNs) have shown to be powerful for this purpose, but the lack of annotated data from these novel devices is a barrier to DNN training. We propose a novel technique based on fine-tuning of linear pre-processing filters, which is capable of compensating for variations in electrode positions and amplifier characteristics and enables training of models for subcutaneous EEG on largely available scalp EEG data. The effectiveness of the method is demonstrated on a state-of-the-art EEG-based sleep scoring model, where we show that the performance on a database used for training can be retained on the subcutaneous EEG by fine-tuning on data from only three subjects.

KW - deep learning

KW - eeg

KW - wearable devices

KW - sleep scoring

UR - https://bsn.embs.org/2023/program/

UR - https://bsn.embs.org/2023/program/session/#1692856189303-4d9f46b4-0d8f

U2 - 10.1109/bsn58485.2023.10331106

DO - 10.1109/bsn58485.2023.10331106

M3 - Conference Proceedings with Poster Presentation

SN - 979-8-3503-3841-6

T3 - International Workshop on Wearable and Implantable Body Sensor Networks (BSN)

SP - 1

EP - 4

BT - 2023 IEEE 19th International Conference on Body Sensor Networks (BSN)

CY - Boston, Massachusetts, USA

T2 - IEEE-EMBS International Conference on Body Sensor Networks – Sensor and Systems for Digital Health

Y2 - 9 October 2023 through 11 October 2023

ER -

Hartmann M , Lechner L, Helge AW, Ahrens E, Bachler M , Hametner B et al. Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models. In 2023 IEEE 19th International Conference on Body Sensor Networks (BSN): Conference Proceedings. Boston, Massachusetts, USA. 2023. p. 1-4. [P48]. (International Workshop on Wearable and Implantable Body Sensor Networks (BSN)). doi: 10.1109/bsn58485.2023.10331106

Fine-tuning of pre-processing filters enables scalp-EEG based training of subcutaneous EEG models

Abstract

Publication series

Conference

Research Field

Keywords

Access to Document

Other files and links

Fingerprint

Cite this