1 of 9

Toward Non-Invasive Depth Recording: Learning Mesial Temporal iEEG from Scalp EEG

Yinuo Qin

Jul 17th 2026

2 of 9

3 of 9

iEEG & Scalp EEG

  • iEEG Pros
    • Direct recording from deep brain structures
    • High spatial resolution (millimeter-scale localization)

​

  • iEEG Cons
    • Highly invasive - requires neurosurgery
    • Risk of hemorrhage and infection
    • Not suitable for healthy participants

4 of 9

Dataset

  • Dataset of human medial temporal lobe neurons, scalp and intracranial EEG during a verbal working memory task (ID: 000574)
  • Data processing: windowing, dropping, train / validation / test split

​

Boran, Ece, et al. "Dataset of human medial temporal lobe neurons, scalp and intracranial EEG during a verbal working memory task." Scientific data 7.1 (2020): 30.

5 of 9

Method

  • Multi-head attention network for individual subject
  • 70/15/15 split for train, validation, and test

6 of 9

Results

  • Validation correlation: 0.317
  • Validation loss: 0.9444
  • Test correlation: 0.310
  • Test loss: 0.8515

7 of 9

Results – Continue

High-performance epoch

Low-performance epoch

8 of 9

Takeaway and Next Steps

  • Takeaways:
    • It is possible to use scalp EEG to predict iEEG
    • Attention models are effective for EEG->iEEG prediction

​

  • Next steps:
    • Cross-subject generalization
    • Multimodal data fusion
    • Improved model architectures

9 of 9

Toward Non-Invasive Depth Recording: Learning Mesial Temporal iEEG from Scalp EEG

Yinuo Qin

Jul 17th 2026