https://github.com/sharmaniraj009/Brain-Tumor-Detection/

Abstract

One of the more serious illnesses that can affect both adults and children is a brain tumor. Eighty to ninety percent of primary tumors of the Central Nervous System (CNS) are brain tumors. Approximately 11,700 people receive a brain tumor diagnosis each year. For those with a cancerous brain or central nervous system tumor, the 5-year survival rate is roughly 34% for men and 36% for women. There are various classifications for brain tumors, including benign, malignant, pituitary, and others. The patients' life expectancy should be increased by using appropriate care, advance planning, and precise diagnosis. Magnetic Resonance Imaging is the most effective method for identifying brain tumors (MRI). The scans produce a massive amount of image data. The radiologist looks over these pictures. Because brain tumors and their characteristics are so complex, a manual examination can be prone to errors.

The use of artificial intelligence (AI) and machine learning (ML) to automate classification techniques has continuously demonstrated higher accuracy than manual classification. Therefore, it would benefit doctors worldwide to propose a system that uses Deep Learning Algorithms to perform detection and classification. These algorithms include Convolution Neural Network (CNN), Artificial Neural Network (ANN), and Transfer Learning (TL).

Objective of the project

The objective of this project is to determine the diagnosis of brain tumor given the MRI. The model can predict up to 3 variations of tumor. The project is built upon the concept of neural networks. A custom made convolutional neural network is implemented to diagnose the tumor.

Convolutional Neural Network

The extended form of artificial neural networks, known as convolutional neural networks (CNNs), is primarily used to extract features from grid-like matrix datasets. For instance, visual datasets with a lot of data patterns, like pictures or videos. The input layer, pooling layer, convolutional layer, and fully connected layers are some of the layers that make up a convolutional neural network. The input image is processed by the Convolutional layer to extract features, the Pooling layer reduces computation by downsampling the image, and the fully connected layer generates the final prediction. The network uses gradient descent and backpropagation to discover the best filters.

Layers

• Conv2D - A tensor of outputs is produced by convolving the layer input with the convolution kernel created by this layer.
• MaxPooling2D - A convolution neural network architecture's max pooling layer is an essential part that lowers the data's dimensions while assisting the neural network in extracting significant features from the input.
• Dense - The feature map that the flatten layer received from the max-pooling layer is transformed into a format that the dense layers can comprehend.
• Dropout - In order to avoid overfitting, a dropout layer in neural networks randomly sets input units to 0 during training. By excluding specific nodes from different training runs, this procedure creates the illusion that they are not a part of the network architecture.
• Dense - A dense layer is one that has strong connections to the layer before it, meaning that all of the neurons in that layer are connected to every other neuron in that layer.

Activation Function

• Relu - Rectified linear units, or ReLUs for short, are non-linear activation functions that are used in deep neural network machine learning. Another name for it is the function that activates rectifiers.
• Softmax - softmax activation function transforms a vector of real numbers into a probability distribution. In a multiclass classification problem, the output layer activation is the softmax function. The softmax function and the sigmoid function are comparable, but the softmax function adds up everything in the raw output in the denominator.

Loss Function

An event or the values of one or more variables are mapped onto a real number that intuitively represents some "cost" connected to the event by a loss function, also known as a cost function. The goal of an optimisation problem is to reduce a loss function.

• Categorical cross entropy - In machine learning, categorical cross-entropy is a frequently utilised loss function, especially in classification tasks where the output variable is a categorical variable with multiple classes. It calculates the discrepancy between the classes' actual probability distribution and their anticipated probability distribution. The formula looks at how well the computer's guesses match the correct answers, and it tries to adjust the computer's guesses so that they become better over time.