Deep Learning Approach for Brain Tumor Classification Implemented in Raspberry Pi

Abstract

Brain tumors are considered the most aggressive diseases, causing death in their high grade. According to the World Health Organization, the mortality rate from brain tumors is increasing. Therefore, treatment and diagnosis are an essential step to improve the survival rate of patients. There are different medical imaging techniques such as Positron Emission Tomography (PET), Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) used to diagnose tumors in brain. In our work we have diagnosed different types of brain tumors using MRI images. The huge amount of data prevents the classification of brain tumors by traditional inspection of MRI images of the patient’s brain, as this method is time-consuming and can lead to errors. This is where Deep Learning comes in. This field has been very successful in recent years. It is developing rapidly to offer promising results in many complex problems such as medical image classification. The Convolutional Neural Networks (CNN) is the most powerful Deep Learning algorithm for automatic medical image classification tasks across multiple layers such as Convolutional layer, Pooling layer and Fully Connected layer. In this paper we focus on the classification of brain tumor MRI images from a dataset that contains 2870 images divided into four classes (Glioma, Meningioma, Pituitary and Normal). We implemented our CNN architecture on Raspberry Pi 4. Raspberry Pi is an embedded device characterized by its low cost and portability. We obtained satisfactory results (89% accuracy), which demonstrates the reliability of our CNN-based brain tumor classification system on embedded devices with limited resources.