/FR-MRInet

Primary LanguagePython

FR-MRInet: An Encoder-Decoder for Brain Tumor Segmentation

FR-MRInet is a deep convolutional encoder used in an auto-encoder that takes an MRI scan of a brain as input and generates an output that highlights the tumor (if present).

Input and Ground Truth

Predictions

Performance in live images

The model takes input images with resolution of 64 * 64 * 1 and generates an output of 24 * 24 * 3. It takes about 600-700 epochs for the model to converge.

Dataset

Link: https://figshare.com/articles/brain_tumor_dataset/1512427

Prerequisites

  1. Tensorflow v1.5 (This version was used in the experiment. It may or may not work with older versions)
  2. TFlearn API
  3. Python 3.x
  4. Numpy
  5. Tensorboard (optional)

Codes

model.py holds the total architectural design of the autoencoder.
main.py is used for training and validating the model.
File loader.py is a helper class that helps to search and load iamges.

Paper Link: herehttps://paperswithcode.com/paper/fr-mrinet-a-deep-convolutional-encoder