Chen, Jianan;
Yang, Huixin;
Xia, Yunjia;
Li, JINCHEN;
Wang, Hanyang;
Carlson, Tom;
Zhao, Hubin;
(2025)
Exploring High-Density DOT for Three-dimensional Imaging of Mental State using CNN and Transformer Models.
In:
Proceedings of the 12th International Conference on Neural Engineering (NER 2025).
IEEE (Institute of Electrical and Electronics Engineers): San Diego, CA, US.
(In press).
Preview |
Text
20250816065533_734992_1354.pdf - Accepted Version Download (635kB) | Preview |
Abstract
Mental fatigue and workload are critical to cognitive health, with relevance in BCIs, neurorehabilitation, and psychiatric care. Accurate decoding can support adaptive interventions and assistive system performance. High-density diffuse optical tomography (HD-DOT), an advanced form of functional near-infrared spectroscopy (fNIRS), offers high spatial resolution for non-invasive monitoring of cerebral hemodynamics, making it well-suited for detecting subtle changes in cognitive states. Deep learning models, particularly Convolutional Neural Networks (CNNs) and Transformers, are highly effective in image-based pattern recognition and thus hold promise for analyzing HD-DOT data. This study investigates the effectiveness of CNN- and Transformer models for classifying mental fatigue and workload using HD-DOT data. We collected data from 16 participants during rest, reaction time, and N-back tasks, and reconstructed 3D images of hemodynamic changes. Both two-class (low vs. high fatigue) and four-class (0,1,2,3-back) classification tasks were performed. CNN models, particularly lightweight architectures like MobileNet, demonstrated strong generalization performance under leaveone-out cross-validation, achieving up to 90.9% accuracy. Transformer models also performed competitively, with MobileViT achieving the highest classification accuracy of 98.6% in the four-class task. These findings highlight the feasibility and effectiveness of combining HD-DOT with lightweight deep learning architectures for accurate and generalizable assessment of cognitive states.
| Type: | Proceedings paper |
|---|---|
| Title: | Exploring High-Density DOT for Three-dimensional Imaging of Mental State using CNN and Transformer Models |
| Event: | 12th International Conference on Neural Engineering (NER 2025) |
| Dates: | 11 Nov 2025 - 14 Nov 2025 |
| Open access status: | An open access version is available from UCL Discovery |
| Publisher version: | https://neuro.embs.org/2025/ |
| Language: | English |
| Additional information: | This version is the author accepted manuscript. For the purpose of open access, the author(s) has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising from this submission |
| Keywords: | Mental fatigue, Mental workload, fNIRS, HDDOT, Deep learning. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Medical Sciences UCL > Provost and Vice Provost Offices > School of Life and Medical Sciences > Faculty of Medical Sciences > Div of Surgery and Interventional Sci |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10219917 |
Archive Staff Only
 |
View Item |
