Decoding Motor Imagery from EEG with Deep Learning
I trained a compact deep-learning model (EEGNet) to read raw EEG brain signals and predict which movement a person was imagining — left hand, right hand, or feet — using the public BCI Competition IV dataset. The model reached roughly 82% accuracy on this motor-imagery task.
~82%
Decoding accuracy (BCI Competition IV-2a)
EEGNet
Compact CNN architecture
4-class
Motor imagery (left hand · right hand · feet · tongue)
9
Subjects evaluated
Results
How it works
Raw 22-channel EEG recordings from the BCI Competition IV dataset (BNCI2014_001) are band-pass filtered and segmented into motor-imagery trials. Each trial is fed into EEGNet — a compact convolutional neural network designed specifically for EEG (Lawhern et al., 2018) — which learns spatial and temporal filters to extract discriminative brain-activity patterns. The network outputs a class prediction for the imagined movement: left hand, right hand, feet, or tongue.
I'm an independent researcher focused on EEG - the electrical signals produced by the brain. My current work uses deep learning to decode motor intention from raw brain activity, and I'm preparing this research for publication.
Alongside it, I'm studying radiologic technology at LaGuardia Community College, training in medical imaging. I care about the same core question in both: how do we read a signal from the human body and turn it into something useful for care?
I work independently, teach myself the tools I need, and document everything rigorously.
Experience & Education
Two parallel paths — reading signals from the brain in the lab, and from the body in the clinic. Same skill, different source.
Work
Independent EEG Researcher
Independent Research
2024 — Present
Self-directed research decoding motor imagery from raw EEG using deep learning. Preparing manuscript for publication.
EEGNet on BCI Competition IV (~82% accuracy)
Python · PyTorch · Braindecode
Signal preprocessing & model evaluation
HR & People Operations
TODO: confirm employer
TODO: confirm dates
People operations, onboarding, and workflow design with a product-minded, systems-thinking approach.
Employee lifecycle & onboarding
Process design & documentation
Cross-functional coordination
Operations & Retail Leadership
TODO: confirm employer
TODO: confirm dates
Store operations, team coordination, and customer-facing leadership in fast-paced environments.
Team leadership & scheduling
Inventory & daily operations
Customer experience
Spatial Computing & Immersive Tech
Independent / Field exploration
2024 — Present
Hands-on exploration of AR/VR platforms, spatial interfaces, and human-computer interaction.
VR headset prototyping & demos
Spatial computing research visits
HCI & interaction design
Education
Radiologic Technology (A.A.S.)
LaGuardia Community College
In progress
2024 — Present
Progress40%
Clinical training in diagnostic imaging — X-ray, patient care, and reading anatomical signals from the body.
Focus areas
Medical imaging
Patient positioning
Radiation safety
Anatomy & physiology
Self-Directed Study — Neuroscience & ML
Independent
In progress
2023 — Present
Progress75%
Structured self-study in brain-computer interfaces, building toward independent research publication.
Focus areas
EEG signal processing
Deep learning for BCI
Motor imagery paradigms
Research methodology
Anatomy & Imaging
In the Lab — From Structure to Signal
Hands-on cranial anatomy study — the physical structure that EEG signals pass through on their way out of the brain.
Skull→Brain tissue→EEG signal
Skull → Brain tissue → EEG signal
Gallery
Photos & activities — tap any photo to view the full image.
Publications
Research output and manuscripts.
Decoding Motor Imagery from EEG Signals Using EEGNet
Tenzin Tsering · Independent Researcher · 2025
A compact EEGNet model trained on the BCI Competition IV motor-imagery dataset achieves approximately 82% decoding accuracy across four imagined-movement classes, demonstrating that deep learning can extract motor intention from raw EEG without hand-crafted features.
