Brain Tumor Detector

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

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

4 Types of scans (of brains):

• T1
• T1ce
• T2
• FLAIR

And the ground truth

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.

Dice score = 2⋅number of true positives / (2⋅number of true positives + number of false positives + number of false negatives)

Algorithm/Technic	Dice Score
SVM (SVC)	0.713
Multi Layer Preceptron	0.744
Random Forest	0.727
CatBoost	0.780

CatBoost seems to win!

PS

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