Decoding the Presence of Tony from sEEG Recordings

Project Overview

In this tutorial, we explore the principles of stereoencephalography (sEEG) data decoding through a practical, hands-on approach. Our objective is a binary classification task: determining the presence of Tony, a character from 'Greenbook', in a movie frame, using sEEG recordings. This project serves as an insightful introduction to the basics of sEEG decoding.

Dataset

Data Access and Setup

The dataset for this tutorial has been pre-processed and is ready for use. Download the dataset using your Brown University email from the links provided in our Slack channel.

Steps to Set Up the Data:

1. Create a Data Folder: In the root directory of this project, create a folder named /data.
2. Download and Organize the Data: Use the links provided in our Slack channel to download the sEEG and label data. After downloading, place these files inside the /data folder.

Methods & Results

Our classification task is approached using two primary machine learning models:

1. Fully Connected Neural Network (FCNN): Achieving a testing accuracy of 80.34% (chance level: 61.19%), with a data split of 70% training, 15% validation, and 15% testing. Below is the training and validation loss graph.
2. Support Vector Machines (SVMs): With the same dataset, this model, using a polynomial kernel and C=0.001, attained the highest accuracy of 67.24%.

Code Organization

The repository is organized into two distinct pipelines: a PyTorch pipeline for the Fully Connected Neural Network (FCNN) and a Scipy pipeline for Support Vector Machines (SVMs).

PyTorch Pipeline for FCNN

• /experiments:

  • binary_label_classification.py: Core training and testing logic for the FCNN model.

• /dataset:

  • /binary_label_dataset.py: Custom dataset class for the binary classification task.

• /eval:

  • /eval.py: Evaluation function used in validation and testing.

• /models:

  • binary_label_fcnn.py: Architecture definition of the FCNN model for the binary classification task.

• /train:

  • train.py: Training procedures specific to the FCNN in PyTorch.

• /utils: Various utility scripts.

  • data.py: Dataset preprocessing management.
  • model.py: Model operation utilities.

Scipy Pipeline for SVMs

• /svm:
  • svm_demo.ipynb: A Jupyter Notebook demonstrating SVM implementation and usage within the Scipy framework.