Brain Tumor Detection Using Deep Learning

Abstract

One of the most leading death causes in the world is brain tumor. Tumor Detection is one of the most difficult tasks in medical image processing. In fact, the manual classification with human-assisted support can be improper prediction and diagnosis shown by medical evidence. The detection task is too difficult to perform because there is a lot of diversity in the images as brain tumors come in different shapes and textures. Recently, deep learning techniques showed promising results towards improving accuracy of detection and classification of brain tumor from magnetic resonance imaging (MRI). In this paper, we propose a deep learning model for the classification of brain tumors from MRI images using convolutional neural network (CNN) based on transfer learning. The implemented system explores a number of CNN architectures, image pre-processing and transfer learning model named MobilNet to achieve the better performance and accuracy.

Introduction

I. INTRODUCTION

 A brain tumor [1] is an abnormal growth or mass of cells in or around the brain. It is also called a central nervous system tumor. Brain tumor can be malignant (cancerous) or benign (not cancerous). Some tumors grow quickly others are slow-growing. Tumor in brain can seriously disrupt the central nervous system. Furthermore, the mass of tumor-cells can affect the brain’s regular functionalities. It should also be noted that many types of tumors make the brain tissue subjected to a scaling-up occurring over time, which leads to brain cells damage. The cause of brain tumors is having exposure to large amounts of radiation from X-rays or previous cancer treatment. Some brain tumors occur when hereditary conditions are passed down among family members. and symptoms of a brain tumor vary depending on the tumor’s location and type, size and what the affected part of the brain controls. It in observed that brain tumors occur more often in men than women. Although they are most common among older adults, they can develop at any age. Brain tumors are the leading cause of cancer-related death in children under age 14. However, early discovery of brain tumors help significantly improve the possibility of treatment and survival rate of the patients. In spite of this, manual classification of tumor using a significant quantity of MRI scans, generated in clinical routine, is time and labor consuming task In fact, the use of the magnetic resonance imaging (MRI) technique in medicine produces high quality images. This kind of imaging is often used by scientists in detecting brain tumors and showing their progress overtime. MRI images play a crucial role in automatic medical analysis field as they facilitate visualizing the different brain structure, thus providing detailed information about it. Scientists have developed different techniques for detecting and classifying brain tumor using MRI images. These approaches range from classical medical image processing to advanced machine learning techniques.

Deep learning [2] is a machine learning technique that teaches computers to do what comes naturally to humans like learn by example. In deep learning, a computer model learns to perform classification tasks directly from images, text, or sound. Deep learning models can achieve state-of-the-art accuracy, sometimes exceeding human-level performance. Deep learning models are trained by using large sets of labeled data and neural network architectures that learn features directly from the data without the need for manual feature extraction. In ML Algorithms everything is flatten and in single dimension array but whereas in deep learning we use some think called as tenser and tenser has basically small matrices inside a big matrix, so it can be consider as matrix nested at inside a matrix. Deep learning is a specialized form of machine learning. A machine learning workflow starts with relevant features being manually extracted from images. The features are then used to create a model that categorizes the objects in the image. With a deep learning workflow, relevant features are automatically extracted from images. In addition, deep learning performs “end-to-end learning” – where a network is given raw data and a task to perform, such as classification, and it learns how to do this automatically. A key advantage of deep learning networks is that they often continue to improve as the size of your data increases. Deep learning has been applied in many applications, such as pattern classification object detection, speech recognition and other decision making tasks. However, the main challenge for DL is the huge data necessary for training.

II. LITERATURE SURVEY

1. Dr. Chinta Someswararao [3], paper titled “Brain Tumor Detection Model from MR Images using Convolutional Neural Network “ was a combination of CNN model classification problem for predicting whether the subject has brain tumor or not & Computer Vision problem for automate the process of brain cropping from MRI scans. The final accuracy achieved by him is much higher than 50% baseline (random guess). However, it could further be increased by larger number of train images or through different models and techniques.
2. Aryan Sagar Methil,[4], paper titled “Brain Tumor Detection using Deep Learning and Image Processing” presents a novel method involving image processing techniques for image manipulation which would aid our CNN model to classify tumor and non-tumor images better. Image Processing techniques helped to solve the illumination issues and brought the tumor into focus. Data augmentation was used to reduce the chances of overfitting, as it artificially expands the size of a training dataset, thus bringing out an improvement in the performance and the ability of the model to generalize. There are limitations to this work as there are small chances that the image pre-processing applied can damage the information which makes a tumor image appear non-tumor in the eye of the CNN model. For future improvements, we can use ensemble techniques and combine the performance of different models for better performance.
3. Sneha Grampurohit [5], paper titled “Brain Tumor Detection Using Deep Learning Models” proposed work in which Deep neural networks such as CNN and VGG-16 are investigated on MRI images of Brain. Both the models have given an effective result, However VGG-16 takes a greater computational time and memory but has given satisfactory results compared to CNN. Due to the availability of huge data being produced and stored by the medical sector, Deep learning will play an important role in data analysis in the upcoming days. 
4. Masoumeh Siar and Mohammad Teshnehlab [6], paper titled “Brain Tumor Detection Using Deep Neural Network and Machine Learning Algorithm” used the combination of feature extraction algorithm and the CNN for tumor detection from brain images is presented. The CNN is capable of detecting a tumor. The CNN is very useful for selecting an auto-feature in medical images. Images collected at the centers were labeled by clinicians, then, tumor screenings were categorized into two normal and patient classes.
5.  Khurram Shahzad and Imran Siddique [7], paper titled “Efficient Brain Tumor Detection Using Image Processing Techniques “ focused on an easy, fully automatic and efficient algorithm for extraction of brain tumor has been introduced. Morphological operation like erosion and dilation along with morphological gradient and threshold are used. Morphological gradient is used for calculating threshold. Threshold is used to binarize the image which results an image having tumor and some noise with it. Erosion is used for thinning the image as it shrinks the image and helps to reduce noise or unwanted small objects. Dilation is being used after erosion so that to get removed tumor portion back which was being removed by erosion.

III. PROPOSED METHOD

The proposed model aims to classify brain MRI images into two classes, images with tumor and images without brain tumor or healthy. First, different preprocessing steps are applied to the MRI images for image augmentation and enhancement. The original dataset consists of MRI images, from which some of them having tumor and some of them are non-tumorous. The datasets are further Split into Train, Validation, and Test sets. The pre-trained CNN architectures are used to test and evaluate the proposed model. The flow chart of the model for detection is given above.

A. Data Acquisition

The dataset used for training and testing was collected from Kaggle. It contains brain MRI images in which some of them are images containing tumor (tumorous images) and some images are normal (without tumor). Tumorous images are segregated in folder named “Brain tumor” and normal images are kept in “healthy” folder. The images are in different formats and of variable sizes.

B. Data Pre-Processing and Augmentation

For any machine learning project data pre-processing is the most crucial and initial step. In this the raw data was collected and making it useful for machine learning model. As mentioned the dataset contains images of different formats and sizes which may contain noise. This can lead to errors in classification and segmentation. Pre-processing the image will definitely reduce this problem and data can be transformed in a standard format acceptable for classification and segmentation. Deep Learning requires large dataset for producing accurate results. Image augmentation is a process of increasing size of the dataset by producing copies of images through different ways of processing like random rotation, shifts, shear and flips by using the ImageDataGenerator tool in Keras TensorFlow. This process boost the model to generalize better and helps prevent overfitting.

C. Design Model

We are using CNN, transfer learning and its architecture called Mobilenet to design our model which are used to improve the accuracy of our model. Before training our model the whole dataset was divided into three parts called as train data and testing data and validation data. Training data is used to train the model and Testing data and validation data was used to test the model. In this project 70% of the data was taken as training data and 15% was taken as testing data as well as validation data.

1. Convolutional Neural Networks (CNN or ConvNet): one of the most popular types of deep neural networks is known as convolutional neural networks A convolutional neural network is a network architecture for deep learning which learns directly from data, eliminating the need for manual feature extraction.CNNs are particularly useful for finding patterns in images to recognize objects, faces, and scenes. In our project [9] The pre-processed image is fed to the CNN model which has a input layer, convolution layers and a fully connected layer which activates a specific neutron to give specific output or decision. The input image forms the input layer. The image is represented as a 224x224 pixel matrix. Each pixel reveals certain features. In the first convolution layer 16 filters of 3x3 size kernels each are applied over the input image by sliding through the position one by one and in total 8 feature maps are produced, this process is called feature extraction. These features are then fed to ReLU activation function which performs a threshold operation to each input element where values less than zero are set to zero. A max pooling layer of 2x2 window size is applied to the output of ReLU layer which results into down-sampling the feature maps. The output of previous convolution layer serves as input to second convolution layer. Second convolution layer consists of 36 filters of 3x3 size kernels which are applied to each of the features maps obtained from previous layer. Similar ReLU and max pooling operations are performed to produce down-sampled data. Same operations are continued for the third and fourth convolution layer where 64 filters and 128 filters of 3x3 size kernels are used. Again ReLU operation is applied and fed to the max pooling layer. The operations performed throughout the fourth layers extracts prominent and important features necessary for accurate classification. The output of the fourth convolutional layer are then flattened to a single matrix and pass to the dense layer. The error between actual result and predicted result is calculated. During training, overfitting can occur when model learns the details and noise in the data used for the training and influences performance of model on new unseen data. To prevent overfitting a dropout function is used in fully-connected layers. Dropout is a regularization technique which randomly drops some units i.e. set to 0 at certain rate. In proposed model 25% of units are dropped to reduce overfitting.
2. Transfer Learning: Transfer learning [10] is a machine learning method where a model developed for a task is reused as the starting point for a model on a second task. It is a popular approach in deep learning where pre-trained models are used as the starting point on computer vision and natural language processing tasks given the vast compute and time resources required to develop neural network models on these problems and from the huge jumps in skill that they provide on related problems. . In this propsed work, we used pre-trained CNNs architectures called MobileNet ,MobileNet [11] is a type of convolutional neural network designed for mobile and embedded vision applications. They are based on a streamlined architecture that uses depthwise separable convolutions to build lightweight deep neural networks that can have low latency for mobile and embedded devices. MobileNet is a class of CNN that was open-sourced by Google, and therefore, this gives us an excellent starting point for training our classifiers that are insanely small and insanely fast.

IV. ACKNOWLEDGMENT

We would like to express our gratitude to our guide Prof. Abdul Razzaque and our Head of Department Prof. Dr. M.S Khatib for giving us agreat opportunity to excel in our learning through this project. We would also like to thank our families and friends for their consistent encouragement throughout the project. This project has helped us to expand our knowledge to a great extend.

Conclusion

In this paper we proposed an efficient method for automatic brain tumor classification using MRI images. The method is based on transfer learning and implemented on well known CNN architectures. Transfer learning has the benefit of decreasing the training time for a neural network model and can result in lower generalization error The time-consuming process of brain tumor detection is thus simplified by automation. An accuracy of about 95% on testing data is achieved by the proposed model for detecting brain tumour. In future work, we can build model of brain tumor for detecting different types of brain tumors and its region and how much percentage of the brain affected through the cancerous cell.

References

[1] Cleveland clinic “Brain Cancer” [February 2020] (online) Available: https://my.clevelandclinic.org/health/diseases/6149-brain-cancer-brain-tumor [2] MathWorks “what is deep learning” [2022] (online) Available: https://www.mathworks.com/discovery/deep-learning.html [3] Dr. Chinta Someswararao “Brain Tumor Detection Model from MR Images using Convolutional Neural Network” IEE May [June 2020] [4] Aryan Sagar Methil “Brain Tumor Detection using Deep Learning and Image Processing “IEE [June 2021] [5] Sneha Grampurohit “BRAIN TUMOR DETECTION USING DEEP LEARNING MODELS”, IEEE [May 2020] [6] Masoumeh Siar “Brain Tumor Detection Using Deep Neural Network and Machine Learning Algorithm” International Conference on Computer and Knowledge Engineering [October 2019] [7] Khurram Shahzad and Imran Siddique “Efficient Brain Tumor Detection Using Image Processing Techniques “International Journal of Scientific & Engineering Research [December 2019] [8] NAVONEEL CHAKRABARTY Kaggle dataset [April 2019] (online) Available: https://www.kaggle.com/datasets/navoneel/brain-mri-images-for-brain-tumor-detection [9] Gajendra Raut “ Deep Learning Approach for Brain Tumor Detection and Segmentation” IEE [May 2021] [10] Jason Brownlee “A Gentle Introduction to Transfer Learning for Deep Learning” [September 2019] (online) Available: https://machinelearningmastery.com/transfer-learning-for-deep-learning/ [11] Abhijeet Pujara “Image Classification with MobileNet” [July 2020] (online) Available: https://medium.com/analytics-vidhya/image-classification-with-mobilenet-cc6fbb2cd470

Copyright

Copyright © 2022 Sufiyan Salim Akbani, Adeeba Naaz, Nazish Kausar, Prof. Abdul Razzaque. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.