Recent work in the Schaffer-Nishimura lab at Cornell has shown a link between "stalls" - clogged blood vessels in the brain - and Alzheimer's, the 7th biggest killer in the world. A citizen science project called Stall Catchers has been identifying stalled vessels in microscope videos taken from the brains of mouse models. Here, I use neural nets attempt to predict whether a vessel is stalled as part of the Clog Loss challenge through DrivenData.
Currently, a ClogData class is implemented that allows retrieval of the train and testing data from an Amazon S3 bucket. Two networks are attempted: both use a CNN (either a custom-trained convolutional autoencoder or pre-trained VGG16 model) to extract features from the series of images of each vessel, and then uses an LSTM in a many-to-one fashion for binary classification. See the jupyter notebooks for examples.