DEEP LEARNING FOR BRAIN TUMOR DETECTION: ADVANCEMENTS AND APPLICATIONS

Abstract

When it comes to medical imaging, brain tumour 
segmentation is a particularly challenging and 
crucial endeavour. This is because manual 
categorization can lead to inaccurate prognosis and 
diagnosis. It's also a lot of effort when there's a lot 
of data to work with. In photographs, it's difficult to 
tell the difference between a brain tumour and 
normal tissue because both have a wide variety of 
appearances and are so similar. Fuzzy C-Means 
was used in this study to remove brain tumours 
from two-dimensional MRI images, which was 
then followed by classical classifiers and also cnn. 
There were tumours of various sizes, shapes and 
intensities included in the dataset used in the study. 
The scikit-learn library's classic classifier used 
SVM, KNN, MLP, LR, Naive Bayes, and Random 
Forest as well as other well-known machine 
learning 
algorithms. 
Convolutional 
Neural 
Networks (CNN) were then examined, which are 
better at predicting outcomes than standard neural 
networks developed with Keras or Tensorflow. Our 
analysis found that CNN's accuracy percentage was 
97.87%. The primary goal of this investigation is to 
use statistical and textural data to discriminate 
between normal and aberrant pixels.