ABSTRACT
A brain tumor is the most common cause for high mortality in children and adults worldwide. A brain biopsy is the standard clinical procedure used to establish diagnosis and decide therapeutic choices in brain tumors located in noneloquent brain areas. Due to its invasive nature, all biopsy procedures carry potential complications such as infection, risk for neurological deficits, and surgical site complications like hematoma. Therefore, developing a noninvasive accurate quantitative assessment of brain tumors is of vital importance and can change the current clinical management of brain tumors. In clinical practice, the manual segmentation of a brain tumor from radiological images is the common procedure. Neuroimaging modalities, especially magnetic resonance imaging (MRI), have shown promise in brain tumor diagnosis and prognostication by assessing tumor type, position, and size noninvasively. MRI is routinely employed in clinical/radiological diagnosis of brain tumors. With the availability of advanced computing facilities, using novel convolutional and deep neural networks, several researchers have developed automatic brain tumor segmentation algorithms that have demonstrated good accuracy in detecting different tumor regions in the brain such as edema, core tumor, and enhancing tumor. A review of these algorithms and a gentle introduction to deep neural network concepts are presented in this work.
