# MI-022: Deep learning-based electroencephalography analysis: a systematic review

> 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-022.md`

## Paper Access

* Internal PDF: <a href={"/papers/MI-022.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: [10.1088/1741-2552/ab260c](https://doi.org/10.1088/1741-2552/ab260c)
* Deployment boundary: these PDF links are intended only for a private/protected internal wiki.

## MI-022: Deep learning-based electroencephalography analysis: a systematic review

## Metadata

* ID: MI-022
* Title: Deep learning-based electroencephalography analysis: a systematic review
* Year: 2019
* DOI / URL: 10.1088/1741-2552/ab260c
* Local PDF: see Paper Access section above
* Text artifact: local-only path withheld from docs site
* Review status: `extracted`

## Study Type

* Track: EEG deep learning / MI context
* Task: systematic review of deep learning for EEG analysis
* Participants or dataset: review of multiple EEG studies and datasets
* Device/electrode setup: varies across reviewed studies
* Protocol/task: multiple EEG domains including BCI, clinical, and cognitive tasks

## Methods

* Signal processing or analysis: systematic coding of architectures, datasets, channels, subjects, tasks, and performance trends
* Metrics: review-level accuracy comparison and study metadata
* Online/offline: literature review

## Key Results

* The review reports a median accuracy gain of 5.4% for deep learning approaches over traditional baselines.
* It warns that EEG datasets are often small and heterogeneous, making generalization difficult.

## Limitations

* Broader than MI; not all included studies are motor imagery or BCI control.
* Review-level gain should not be treated as a guarantee for any specific project.
* Need to cite specific primary papers for concrete decoder claims.

## Relevance To Current Review

* Supports trend discussion around deep EEG models while preserving caution.
* Useful for avoiding overclaiming deep learning as a universal solution.

## Evidence Status

| Claim | Status | Evidence Note |
| --- | --- | --- |
| Deep learning is a major EEG analysis trend, but dataset size/generalization remain problems. | verified | Abstract and introduction discuss promise and data limitations. |
| Review-level median gain was 5.4%. | verified | Abstract reports the median gain. |
| Deep learning should replace all classical MI baselines. | needs confirmation | Review does not support dropping CSP/FBCSP/Riemannian baselines. |

## Open Questions

* Which deep-learning claims require primary MI papers rather than this broad review?
* Should deep models be treated as P1 after classical baselines in experiments?