Magnetic resonance imaging-based brain tumor grades classification and grading via convolutional neural networks and genetic algorithms

Abstract

Gliomas are the most common type of primary brain tumors in adults and their early detection is of great importance. In this paper, a method based on convolutional neural networks (CNNs) and genetic algorithm (GA) is proposed in order to noninvasively classify different grades of Glioma using magnetic resonance imaging (MRI). In the proposed method, the architecture (structure) of the CNN is evolved using GA, unlike existing methods of selecting a deep neural network architecture which are usually based on trial and error or by adopting predefined common structures. Furthermore, to decrease the variance of prediction error, bagging as an ensemble algorithm is utilized on the best model evolved by the GA. To briefly mention the results, in one case study, 90.9 percent accuracy for classifying three Glioma grades was obtained. In another case study, Glioma, Meningioma, and Pituitary tumor types were classified with 94.2 percent accuracy. The results reveal the effectiveness of the proposed method in classifying brain tumor via MRI images. Due to the flexible nature of the method, it can be readily used in practice for assisting the doctor to diagnose brain tumors in an early stage.

Introduction

Brain tumor referred to the aggregation of abnormal cells in some tissues of the brain. According to the brain tumors origin, they are divided into two categories, primary and metastatic brain tumors. The origin of primary brain tumors is in the brain, while metastatic brain tumors originate from other body parts. Tumors can be cancerous (or malignant) or noncancerous (or benign). Malignant brain tumors grow fast and spread to other areas of the brain and spine and compared to benign tumors, they are more life-threatening. A more detailed categorization classifies tumors into four grades where, the higher the grade, the tumor is more malignant. Due to the presence of brain tumors in the center of the nervous system of human body, even benign tumors may incapacitate the brain and cause irrecoverable effects.

Gliomas are considered as the most common type of primary brain tumor in adults [1]. According to the World Health Organization grading system [2], gliomas are diagnosed in grades of severity from I to IV. Grade I tumors have cells that are benign and are approximately normal in appearance. Grade II tumors have cells that appear to be slightly abnormal. Grade III tumors have cells that are malignant and clearly abnormal. The most severe type of brain tumors that contain fast-spreading and abnormal cells are considered as Grade IV. Glioblastoma multiforme (GBM) are quintessential tumors of this type. Meningioma tumors, arise from a layer of tissue called the meninges. Meninges cover the brain and spinal cord and act as protector. They are mostly considered as benign tumors, because they grow at a slow pace and are also less likely to spread. Pituitary tumors develop in the pituitary gland and account for 14% of all primary intracerebral tumors, with most of them are due to spontaneous mutation and some are due to inherited genetic defects [3]. These tumors are also benign and they are much less likely to spread. Although these tumors are considered benign, they can cause serious health problems due to their presence in sensitive areas of the brain [4].

Early detection plays a major role in treatment and recovery of the patient [5]. Diagnosing a brain tumor and its grade usually undergoes a complicated and time-consuming process. Usually, the patient refers to MRI when the brain tumor has grown sufficiently and several harassing symptoms have appeared. After examining the brain images, if tumor existence is suspected, the patient's brain biopsy comes to action. Unlike MR, biopsy has an invasive procedure and in some cases, it may even take up to a month for a definite answer. MRI specialists perform techniques such as perfusion to grade tumor and biopsy to confirm. It should be noted that in recent years some novel methods have been introduced in order to grade brain tumors other than biopsy. In particular, distinguishing high-grade and low-grade glioma using perfusion MR imaging has been able to resolve some biopsy drawbacks. For these reasons, utilization of a computer-aided system for detection is helpful. An automatic efficient system for brain tumor classification assists doctors in interpretation of medical images and supports decision of specialists in an early stages of tumor growth. In this study, brain tumor grading is done by spending much less time and as is confirmed in this paper with high accuracy. Furthermore, the whole process of classification is non-invasive.

Considerable attention has been paid to medical image analysis for diagnosis purposes. Recently, the emergence of modern machine learning algorithms and their proven efficiency in solving various problems in the field of artificial intelligence, have also doubled the interest to the field of health-related topics and algorithms [6]. Many researches on classifying various tumors using MRI, especially MR brain images, artificial neural networks and evolutionary algorithms have been done [7], [8], [9] and various methods have been implemented as well. Previous studies indicate that normal and abnormal classes in brain MR images are easily distinguished using shallow machine learning algorithms such as support vector machine, neural network, Hybrid intelligent techniques and probabilistic neural network [10], [11], [12], [13], [14].

In [15] classification schemes and their performance in order to classify several types of brain tumors and grades of Gliomas are investigated. In their proposed method, the region of interest is first defined and then some features such as the shape of the tumor are extracted from the MR images. In order to select the appropriate features, support vector machines (SVM) with recursive feature elimination has been used. By observing their results, it is seen that their proposed method in binary classifications has obtained high accuracies, but the accuracy of the multi-class classification according to the confusion matrix provided in this paper, is low. In a recent article, [16], the classification performance of three tumor types using fully connected and convolutional neural networks (CNNs) is compared. As described in this article, the performance of various structures of the convolutional networks has been tested and, eventually, a relatively shallow network with two convolutional layers, two max-pooling layers, and two fully connected layers is used for classification. It has also been mentioned that the use of Vanilla preprocessing has been effective in classification accuracy. In another recent study [17], CNN is used to classify healthy and unhealthy brain images as well as high-grade and low-grade Glioma tumors. A modified version of the famous AlexNet was used as their network architecture. Despite the valuable works being done in this area, developing a robust and practical method to classify brain MR images still requires more effort.

Convolutional neural networks have had many remarkable successes in solving complex problems of machine learning and currently are considered as the most successful method for image processing [18]. Instead of matrix multiplication, convolution operators are used in most layers of these networks. This contributes to the superiority of convolutional networks in solving problems with high computational costs. This is very important since the MRI datasets in MRI-based diagnosis include thousands of images with different qualities and types. Another advantage of this method compared to shallow machine learning methods, is automatic feature extraction. In conventional methods, a method was usually proposed for extracting features, and to further reduce the dimensions, a method was used to select the dominant features. Recently, CNN has also been widely used in the processing of medical images using deep neural networks such as grade classification [19], segmentation [20], [21] and skull stripping of brain tumor images [22].

In this article, a method based on CNN is proposed to classify three grades of Gliomas with MR images. Selecting an appropriate deep neural network architecture for a specific purpose, consists of a challenging procedure which is usually done by trial and error or employing a common architecture. Unlike conventional schemes, in the proposed method the architecture of the convolutional neural network is evolved using GA. Networks with different number of layers and parameters are investigated by GA and the network with the best performance on the dataset was selected for further processing. Afterwards, a model averaging method called bagging is utilized on the best model evolved by the GA. Bagging is an ensemble method and is employed in order to decrease the variance of final diagnosis. The proposed method is used in two case studies. In first case, three Glioma grades are classified with 90.9% accuracy. In second case, for demonstrating the strength of the proposed method, three different types of tumors from another MRI database were used as the input to CNN, and the final performance of diagnosis was 94.2%. Results confirm that proposed method is applicable on different brain MRI datasets in order to assist the specialist in early detection.

The remainder of the paper is organized as follows. In Section 2, a brief explanation about the datasets utilized as input of the networks is given. CNNs are discussed in details in Section 3. In Section 4, the proposed method for selecting an appropriate architecture based on GA is presented. Experimental results are presented and discussed in Sections 5 Results, 6 Discussion, respectively. Finally, Section 7 is dedicated to conclusions.