/BrainTumorsDetector-T34

:four_leaf_clover: Brain Tumors Detector with Machine Learning on BRATS18 Dataset :dizzy:

Primary LanguageJupyter Notebook

Brain Tumor Detector

Machine Learning algorithm to detect tumors in medical images of brain slices.

Brain tumors

Samples_Dataset_GD_1

Requirements

  • Python 3

  • Linux (other OS were not tested)

  • Python packages required are in requirements.txt, install them with :

pip3 install --user -r requirements.txt

Usage

jupyter-notebook BrainTumorsDetector.ipynb

DataSet

Using the BRATS18 dataset form the (Multimodal Brain Tumor Segmentation Challenge 2018)[https://www.med.upenn.edu/sbia/brats2018.html]

4 Types of scans (of brains):

  • T1
  • T1ce
  • T2
  • FLAIR

And the ground truth

Samples_Dataset_GD_0

Result of different algorithms

  • My metric is the dice score, unstead of computing a score on raw pixels, because there is more zones without tumors than with tumors.
Algorithm/Technic Dice Score
SVM (SVC) 0.713
Multi Layer Preceptron 0.744
Random Forest 0.727
CatBoost 0.780

CatBoost seems to win!

Prediction_with_CatBoost

PS

T-34 is for the Russian tank model, because tanks are cool !