Lesion Growth Predictor is a course project in development which predicts lesion growth for patients with stroke. Read about our findings here.
Please contact Fabien Scalzo to access the data used for this project. Specifically ask for the matrices.
The dataset used in this project (not included in this repository) contains data associated with perfusion imaging of the brains of anonymous stroke patients admitted to the University of California, Los Angeles Medical Center :hospital:. The data was pre-processed (via image segmentation, registration, etc.) by Scalzo's lab team.
Functionality has been implemented to load the data from the dataset (not included). Tools to visualize the data have also been created to look for any funny values or outliers, examine a range of values more closely, and figure out which features seem important. In addition, methods to help tune hyperparameters and record results have been fully implemented.
Incremental/online machine learning algorithms are being explored to see how well it performs in predicting perfusion parameters of interest. In particular, stochastic gradient descent and the Passive-Aggressive algorithms will be evaluated.
Download or clone this repository.
Make sure Python 2.x is installed on your machine. The program currently does not support Python 3. The following packages are used in the scripts.
Note that there is no dataset published to this repository, so make sure a data set is included as a subdirectory in this repository before running this project. The base class DataSet can also be implemented in data.py to provide an interface to extract the data from the dataset of your choice. If the dataset being used belongs to Scalzo, make sure that the path to the datasets indicated in data.py is modified accordingly. The included packages in data.py should also be verified. The main script may also need to be modified accordingly.
Then, open up a terminal (or command line), change directories to this repository directory, and enter the following to run the script to evaluate the machine learning techniques.
$ python main.py
If you are utilizing the dataset from Scalzo's team, then that's it! Just follow the instructions from the command prompt. It's user-friendly!