🫐 Lemons to Lemonade

Latest Notes

Epileptic Seizure Detection with Tiny Machine Learning

3 min read

The rest of this page is still being written

A project for my BSc Computer Science dissertation which achieved a final grade of 93%. The project was supervised by Dr Nick Pham.

This post and GitHub repository may continue to receive updates.

  • View the dissertation report: LINK
  • The notebooks used in the study is on the public GitHub repo
    • (The review is overdue a cleanup)

Aims and Objectives

The main aim of this study is to develop Tiny Machine Learning models for seizure detection. It is split into 3 objectives.

  • An informative feature extracted technique
  • Reliable machine learning model(s) for three types of seizures
  • An Arduino application for inference representing a wearable device

Three types of seizures are to be detected: absence seizure, tonic-clonic seizure, and generalised non-specific seizure.

Methods

Following the 3 objectives, the study is split into 3 parts.

Preparing EEG Data

  1. EEG data is converted into melspectrograms to capture the spatial and temporal data of EEG windows.
  2. Three datasets are produced, one for each seizure
  3. Datasets are balanced with SMOTE

CNN Models

3 individual models, and 1 multi-class classification model, is trained on the produced datasets with a small CNN classifier.

The combined model makes use of all 3 trained individual models within its architecture. This approach provides a direction for multi-class classification of the problem. However, the focus of the study remained on the binary classification tasks.

Each binary classification model produced reliable results.

Arduino Deployment

The PyTorch model is converted to ONNX, and then Tensorflow, before being quantized and deployed to an Arduino platform via TFLite.

A TFLite inference environment was setup on the platform, and inference was successfully conducted.

Achievements

Publications

  1. The preliminary results of this study was accepted into the 2024 MobiUK Wearable and Ubiquitous Systems Research Symposium and was presented:
Epileptic seizure detection with Tiny Machine Learning
Loic Lorente Lemoine, Nhat Pham, MobiUK 2024 - Sixth UK Mobile, Wearable and Ubiquitous Systems Research Symposium

View the abstract: LINK View the presentation slides: LINK

Awards

  1. A short poster featuring the results of the study was submitted to a Cardiff University and Vietnam National University Student Poster Competition on “AI, Smart Healthcare, and IoT” and won first prize:
First Prize, Student Poster Competition on “AI, Smart Healthcare, and IoT”
Global Wales, Cardiff University and Vietnam National University - Ho Chi Minh University of Technology

View the poster: LINK

  1. I was presented with the “Best Final Year Project” award at my graduation:
Best Final Year Project, BSc Computer Science and Variants
Cardiff University, School of Computer Science and Informatics

Graph View