# MI-021: An End-to-end Deep Learning Approach to MI-EEG Signal Classification for BCIs

> Generated from a local paper card. Do not edit this page directly; edit the source card and rerun `vp run docs:generate`.

> Internal wiki boundary: paper-card evidence and user-provided PDF downloads are available for private reading. Extracted full-paper text is not published.

> Source: `library/paper_cards/MI-021.md`

## Paper Access

* Internal PDF: <a href={"/papers/MI-021.pdf"} download style={{ display: "inline-flex", alignItems: "center", justifyContent: "center", minHeight: "2.25rem", padding: "0.45rem 0.8rem", borderRadius: "6px", backgroundColor: "#047857", color: "#ffffff", fontWeight: 700, lineHeight: 1, textDecoration: "none", boxShadow: "0 1px 2px rgba(15, 23, 42, 0.22)" }}>Download Paper</a>
* DOI / official page: [10.1016/j.eswa.2018.08.031](https://doi.org/10.1016/j.eswa.2018.08.031)
* Deployment boundary: these PDF links are intended only for a private/protected internal wiki.

## MI-021: An End-to-end Deep Learning Approach to MI-EEG Signal Classification for BCIs

## Metadata

* ID: MI-021
* Title: An End-to-end Deep Learning Approach to MI-EEG Signal Classification for BCIs
* Year: 2018
* DOI / URL: 10.1016/j.eswa.2018.08.031
* Local PDF: see Paper Access section above
* Text artifact: local-only path withheld from docs site
* Review status: `extracted`

## Study Type

* Track: MI
* Task: end-to-end CNN for MI EEG classification
* Participants or dataset: PhysioNet MI dataset; local text uses 109 subjects and a selected subset of 105 subjects
* Device/electrode setup: 64 EEG channels sampled at 160 Hz
* Protocol/task: imagined movements from PhysioNet motor imagery recordings

## Methods

* Signal processing or analysis: global CNN classifier and transfer learning adaptation for individuals
* Training/calibration: global model trained across subjects, then adapted to single individuals
* Online/offline: offline dataset analysis

## Key Results

* Selected global classifier reached mean accuracies of 80.38%, 69.82%, and 58.58% for two-, three-, and four-class datasets.
* Transfer learning improved overall mean accuracy when adapting the global classifier to individuals.

## Limitations

* Uses a large public dataset rather than the project's low-channel setup.
* Offline performance does not establish online no-control safety.
* Accepted manuscript status should be noted for redistribution review.

## Relevance To Current Review

* Important for deep end-to-end MI classifier trend.
* Supports using transfer adaptation rather than assuming one global decoder will work for all users.

## Evidence Status

| Claim | Status | Evidence Note |
| --- | --- | --- |
| End-to-end CNN MI classifiers can be trained on large public MI datasets. | verified | Methods describe PhysioNet data with 109 subjects and 64 channels. |
| Individual adaptation improved the global classifier. | verified | Abstract states transfer learning improved overall mean accuracy. |
| End-to-end deep MI is ready for project online intervention commands. | needs confirmation | No online no-control or robotic command validation. |

## Open Questions

* Could PhysioNet pretraining help the project if local MI data are scarce?
* What performance drop should be expected with fewer than 64 channels?