SqueezeNet Deep Learning Model for Magnetic Resonance Imaging Brain Tumor Detection

Abstract

The growth of cells in an abnormal and uncontrolled manner leads to tumors. Tumors can be either benign or cancerous. The tenth highest trigger of mortality for both women and men is cancer of the brain and other nerve-related systems. Despite the multiple efforts and accomplishments in brain tumor detection and treatment, exact segmentation and classification continue to be difficult. This is because brain tumors can fluctuate in size, form, and position, and diagnosing one can be quite difficult. Multiclass classification of brain tumors is a consequential research area in Medical Imaging. Brain tumor imaging research has expanded significantly in recent years. More studies have been done on the use of Magnetic Resonance Imaging (MRI). This study focuses on MRI brain tumor detection using Squeezenet in its default values for extracting informative features and classifying the features by employing machine learning (ML) classifiers of the relevant features from Squeezenet using the orange data mining tool after preprocessing. The efficacy of the model was evaluated utilizing 3264 MRI brain tumor datasets comprising glioma, pituitary, meningiomas, and no tumor. Several metrics were utilized to assess the model’s performance with Artificial Neural Networks (ANN) outperforming all the other classifiers with an accuracy of 91%. This study will enable the physician to detect, diagnose, and treat brain tumors at the early stages thereby reducing death and increasing the survival rate of an individual affected by this disease.