← Projects

Clinical EEG Alert System

A client/server system that routes EEG data from bedside to database to a physician, with a 3D electrode-review GUI at the end of the chain.

• Python
• Flask
• Tkinter
• MNE

View source on GitHub

A graduate systems-design project (Duke BME 547, built with a partner) built around a realistic clinical workflow rather than a toy CRUD app: get EEG data from a bedside device to the physician who needs to act on it, with a human in the loop at every handoff.

The end-to-end workflow: a seizure-classified EEG trace triggers an alert that fans out to the database, the nurse's station, and an email to the attending — who opens a server-rendered 3D electrode-review GUI.

The pipeline

1. A Tkinter client posts EEG data (name, patient ID/MRN, heart rate, and the ECG trace itself) to a Flask server as JSON.
2. The server stores it in an in-memory database keyed on MRN — new data for an existing MRN overwrites in place — and raises an alert.
3. That alert is pushed to a nurse’s station GUI, where the nurse confirms they’ve attended to the patient and logs qualitative notes back to the server.
4. The server emails the attending physician a link to a server-rendered page showing a 3D rendering of the brain for identifying the significant electrode contacts, built on MNE’s electrode-localization tooling. The physician’s electrode picks are POSTed back and stored against that patient’s record for later seizure-focus review.

Server API

A small REST surface carries the whole workflow: POST /new_patient upserts a patient’s data (only patient_id is required); GET /get returns every patient on file; GET /get/<mrn_or_name> looks up one patient by MRN or name (falling back to the most recent MRN on a name collision); GET /get/<mrn_or_name>/image renders the stored ECG trace as an HTML page. The database itself is a small class wrapping a list of dictionaries, with per-key attribute lists kept in sync as entries are appended — enough structure for fast lookups without pulling in a real database engine for a project of this scope.

Client-side

The Tkinter client reads a local ECG CSV, plots it with matplotlib, and computes heart rate from the trace before sending everything to the server. The same client pulls existing patient records back down from a dropdown of MRNs already on the server, so the “browse a local file” and “look up what’s already stored” paths share one interface.