In this project I used a dataset of brain MRI images. This dataset contained two classes, one class belongs to images which are healthy brains. The other class belongs to images of brains with tumor.
I dowloaded the dataset from link bellow:
adress: https://www.kaggle.com/navoneel/brain-mri-images-for-brain-tumor-detection
The class of healthy brains contained 154 images.
The class of images with tumor contained 155 images.
I used a CNN neural network to make a binary classifier. At first, I trained the model for 10 epochs:
and I fined tuned it for 30 more epochs.
I reached the value accuracy=0.92
#Model with Augmented Images
I augmented images by:
- rotating
- width and height shifting
- horizontal fliping
- shearing
- zooming
An example of this augmentation is placed bellow:
Then I repeated the model training:
I used a pretrained model, VGG16, I freezed it and then I added one dense layer and a sigmoid classifier layer. I trined it for 100 epochs.
The best achieved accuracy for validation during training was 0.94