How It Works
Neuralitx runs your neural recording through a five-stage pipeline — signal processing, artifact detection, literature retrieval, and AI interpretation — in a single upload.
Upload your EEG or LFP recording in EDF format. EDF is the standard format exported by most clinical and research recording devices including Neuralynx, BrainProducts, and Grass Technologies.
No preprocessing required. Neuralitx accepts raw recordings directly from your device.
The recording is passed through a Python signal processing pipeline. A bandpass filter (1–40 Hz) removes low-frequency drift and high-frequency noise. Frequency band power is then extracted across five canonical bands: delta, theta, alpha, beta, and gamma.
Artifact detection uses amplitude thresholding to flag segments where signal amplitude exceeds three standard deviations from the mean.
A feedforward neural network trained on EEG frequency band features classifies the signal as clean or artifacted. The network outputs a confidence score alongside its classification.
The classifier uses the five extracted band power values as input features, normalized to a 0–1 range before inference.
Before generating an explanation, Neuralitx queries a vector database built on neuroscience literature. Based on the dominant frequency band and signal quality, relevant context is retrieved — what that band pattern typically indicates, what artifact levels suggest, and what the research says.
This grounds the AI explanation in real neuroscience rather than generic language model output.
The processed signal data, neural network output, and retrieved literature context are passed to Claude. The model generates a structured plain English report covering signal quality, dominant frequency patterns, notable findings, and recommendations for the researcher.
The explanation is grounded in your actual data — not a generic summary.
Free during beta. No account required. Upload your EDF file and get results in under a minute.
Try Neuralitx Beta