Dataset

Brain Mapping:

• Datasets are obtained through google image scraping and from here
• For mask images, JSON files are created by VGG Image Annotator and then converted to png file.

Tumor Detection & Classification:

• Datasets are obtained from here

Data Augmentation:

• Purpose: the massive increase in the scale of datasets that will be used for training/validation/test process for CNN model and U-Net architecture.
• Transformed versions of the original image dataset are created by applying image flip/rotation, changing image contrast, etc.

Brain Mapping

Input Validation (Image Classification):

• User input is validified through image classification using the CNN model. Only a valid type of image, sagittal view of brain MRI, can be used for brain mapping.

Brain Mapping (Image Segmentation):

• U-Net architecture is used for brain mapping. The U-Net architecture was inspired by U-Net:Convolutional Networks for Biomedical Image Segmentation. Between each layer, the size of the layer is reduced to avoid overflow.

Tumor Detection & Classification

Input Validation (Image Classification):

• User input is validified through image classification using the CNN model. Only a valid type of image, brain MRI (it can be any view), can be used for brain mapping.

Tumor Detection & Classification:

• Through CNN model image classification, the given image can be classified as one of the following: no tumor, glioma, meningioma, and pituitary.

Web Application

• Application is created using Flask and it is deployed on AWS (tba maybe Heroku??).