Accurate brain tumor detection using deep convolutional neural network

Md Saikat Islam Khan; Anichur Rahman; Tanoy Debnath; Md Razaul Karim; Mostofa Kamal Nasir; Shahab S Band; Amir Mosavi; Iman Dehzangi

doi:10.1016/j.csbj.2022.08.039

Accurate brain tumor detection using deep convolutional neural network

Comput Struct Biotechnol J. 2022 Aug 27:20:4733-4745. doi: 10.1016/j.csbj.2022.08.039. eCollection 2022.

Authors

Md Saikat Islam Khan¹, Anichur Rahman^{1

2}, Tanoy Debnath^{1

3}, Md Razaul Karim¹, Mostofa Kamal Nasir¹, Shahab S Band⁴, Amir Mosavi^{5

6}, Iman Dehzangi^{7

8}

Affiliations

¹ Department of CSE, Mawlana Bhashani Science and Technology University, Tangail, Bangladesh.
² Department of CSE, National Institute of Textile Engineering and Research (NITER), Constituent Institute of the University of Dhaka, Savar, Dhaka 1350, Bangladesh.
³ Department of CSE, Green University of Bangladesh, 220/D, Begum Rokeya Sarani, Dhaka 1207, Bangladesh.
⁴ Future Technology Research Center, College of Future, National Yunlin University of Science and Technology, 123 University Road, Section 3, Douliou, Yunlin 64002, Taiwan.
⁵ Institute of Information Engineering, Automation and Mathematics, Slovak University of Technology in Bratislava, Bratislava, Slovakia.
⁶ John von Neumann Faculty of Informatics, Obuda University, 1034 Budapest, Hungary.
⁷ Center for Computational and Integrative Biology, Rutgers University, Camden, NJ 08102, USA.
⁸ Department of Computer Science, Rutgers University, Camden, NJ 08102, USA.

Abstract

Detection and Classification of a brain tumor is an important step to better understanding its mechanism. Magnetic Reasoning Imaging (MRI) is an experimental medical imaging technique that helps the radiologist find the tumor region. However, it is a time taking process and requires expertise to test the MRI images, manually. Nowadays, the advancement of Computer-assisted Diagnosis (CAD), machine learning, and deep learning in specific allow the radiologist to more reliably identify brain tumors. The traditional machine learning methods used to tackle this problem require a handcrafted feature for classification purposes. Whereas deep learning methods can be designed in a way to not require any handcrafted feature extraction while achieving accurate classification results. This paper proposes two deep learning models to identify both binary (normal and abnormal) and multiclass (meningioma, glioma, and pituitary) brain tumors. We use two publicly available datasets that include 3064 and 152 MRI images, respectively. To build our models, we first apply a 23-layers convolution neural network (CNN) to the first dataset since there is a large number of MRI images for the training purpose. However, when dealing with limited volumes of data, which is the case in the second dataset, our proposed "23-layers CNN" architecture faces overfitting problem. To address this issue, we use transfer learning and combine VGG16 architecture along with the reflection of our proposed "23 layers CNN" architecture. Finally, we compare our proposed models with those reported in the literature. Our experimental results indicate that our models achieve up to 97.8% and 100% classification accuracy for our employed datasets, respectively, exceeding all other state-of-the-art models. Our proposed models, employed datasets, and all the source codes are publicly available at: (https://github.com/saikat15010/Brain-Tumor-Detection).

Keywords: Brain tumor; Computer-assisted diagnosis; Convolutional neural network; Data augmentation; Magnetic reasoning imaging.