Article Preview
Top1. Introduction
A brain tumor is the deadliest diseases of the world that affects life of human kind of all ages, genders, and ethnicities. A brain tumor is an abnormal development of irregular tissue in the brain that can be a tumor or non-tumor (Kumar et al. 2021). For a neurologist, employing the medical assistant tool such as Computer-Aided Diagnosis (CAD) for brain tumor analysis, categorization, and detection are significant concerns. There are mostly three common classes of brain tumor such as meningioma, glioma, and pituitary. For Brain cancer treatment, a precise and prompt diagnosis is required to provide an effective therapy. Treatment options are determined by the pathological kind, the phase of the tumor at the time of evaluation, and the tumor's grade. CAD technologies have aided neurologists in a variety of ways such as in the grading, categorization, and identification of tumors presented by Deepak, S. and Ameer, P. M. (2019). The majority of brain tumor identification is focused on developing an action plan for the segmentation of any specified region. Unfortunately, integrated feature extraction and classification problem has received less attention, although it is not only the most significant phase but also has the potential to develop the performance of CAD for medical images.
Recently, researchers have focused on deep learning approaches for improving the performance of computer-aided diagnosis for medical in the investigation of brain tumor malignancy (Noreen et al. 2020). Brain tumors are classified as cancerous or noncancerous. Noncancerous is benign, whereas cancerous is malignant. Gliomas and meningioma are low-grade tumors that might be classified as benign, as shown in Fig.1. High-grade tumors, such as astrocytoma and glioblastoma, are classified as malignant. Radiologists utilize several imaging methods to identify brain cancers, including PET, SPET, CT, and MRI. Magnetic Resonance Imaging (MRI) is the best system for identifying tumors. This type of tumor requires further investigation to improve tumor detection in the human brain. The classification used deep learning neural networks (DLNN) to classify the various types of tumors after the segmentation of a specific region of a given input image. In the segmentation process, finding the pixel is a difficult task. The segmentation used input images for the neural network after normal and abnormal correct classification regions were present in the MR brain slices. T1, Flair, T1c, and T2 are some of the MRI sequences used to identify brain tumors. The high-quality brain tumor images collected from BRATS (Multimodal Brain Tumor Image Segmentation Benchmark) include High-Grade Glioma (HGG) and Low-Grade Glioma (LGG) respectively (Gupta et al. 2018).
This paper has used figshare datasets of the brain tumor, which are publicly available (J. Cheng 2019). The infiltrative tumors are gliomas with fuzzy borders, making them difficult to spot in images of different intensities. It tackles the challenge of variable intensity stages due to varied MRI machine setups. Many recent techniques employed four modalities T1, T1c, T2, and FLAIR to provide unique and essential data relating to each area of the tumor (Ranjbarzadeh et al. 2021). Multiple MRI modalities have been developed to increase the retrieval from the database of MR images. These modalities are T1, T2, where T1_weighted from MRI with T2_weighted and contrast development (T1c), each of which provides different types of information about tumor pixels presented by Abbasi, S. and Pour, F. T. (2014). For the same tumor area, several modalities give varied intensity levels and patterns. This makes establishing an interaction between the various tumor classifications in the image much easier. For the identical tumor area, modalities of T1 and T1c produce contrast-enhanced, brighter images are modalities that produce dark-shaded images of T2 and T2flair.
Figure 1.
The brain tumor MRI sequences slices areGlioma, Meningioma, and Pituitarydataset samples for three classes