Brain AVM Classification Project

Dataset

Dataset Details

This dataset contains 6,684 images of human brain MRI images which are classified into 2 classes:

• avm
• non-avm

Data Pre-processing

After splitting them into train-val-test datasets with 70-20-10 ratio, we had

• 4678 for training
• 1338 for validation
• 668 for testing

Augmentation

After cropping our images, augmentation was needed to improve our accuracy. We used rotation, horizontal and vertical flip, brigthness change, shear, zooming as techniques.

Model

Pre-trained Model Selection

To perform image classification with higher accuracy, we used pre-trained models from keras's application module. VGG16, VGG19, ResNet50v2, MobileNetv2 and EfficientNetB0 are the pre-trained models that we trained and compared by their accuracy.