Brain tumors are a type of disease that affects specific parts of the brain and are caused by abnormal brain cells. The two types of brain tumors that can be identified are benign(nonharmful) and malignant(harmful)(Abiwinanda et al. 2019; Brindha et al. 2021). There are four distinct grades of brain tumors: grades I and II are called "low grade gliomas" while grades III and IV are called "high grade gliomas." Malignant tumors, also known as Grade III or IV tumors, grow quickly and must be identified as soon as possible(Özyurt et al. 2019). The most common type of malignant brain tumor is glioma which grows in the glial cells of the brain has the potential to invade nearby healthy tissues and has immediate side effects such as decreased life expectancy(Ayadi et al. 2021). Meningioma’s, on the other hand, do not spread rapidly and may be surgically removed, however diagnosing them early on might be challenging. Pituitary tumors are benign tumors that affect the pituitary glands and do not spread to other regions of the body. Loss of eyesight and hormone deficiencies can result from pituitary tumors(Ayadi et al. 2021). When diagnosing and treating brain tumors, early detection, accurate grading, and categorization are essential. Throughout the history of medical image analysis, different techniques have been employed to identify and categorize brain tumors. Many medical experts use a variety of MR images to design effective treatments for diagnosing brain tumors (Gu et al. 2021). Brain tumors are currently diagnosed and classified via histological analysis of tissue samples.Radiologists use MRS metabolite ratios to assess tumor grades manually. Nonetheless, this treatment is unique because radiologists evaluate the reports manually using the metabolites peaks ratios which allows a possibility of human error. These limitations make it imperative to develop a completely automated method for identifying and classifying brain tumors(Irmak 2021). In recent years, brain tumor classification has been addressed by medical professionals through the use of automated computer-aided techniques. Many machine learning and deep learning techniques have been used by previous researchers. Many models, including DenseNeT, ResNET, Google Net, VSM, and Inception V3, have been used for classification tasks. Milica M. Badza et al. classified of brain tumors (meningioma, glioma and pituitary tumors) using convolutional neural network (CNNS) 10 fold cross-validation methods. He used pretrained models on T1-weighted contrast-enhanced magnetic resonance images. He achieved 96.56% accuracy on training data(Badža and Barjaktarović 2020). Ayadi et al. used computer-assisted diagnosis for the classification of brain tumors into different grades. They ran three different types of MR images into a model and achieved 94.74% accuracy (Ayadi et al. 2021).Pereira et al. used CNN models for the visualization of glioblastoma multiforme. Using this approach, he was able to make predictions directly from MRI scans, eliminating the requirement for a region of interest. He had a grade prediction from the entire brain of 85% and a tumor ROI of 92.8%(Pereira et al. 2018). Abiwinanda et al. proposed CNN models to classify three different types of brain tumors (meningioma, pituitary and glioma tumors). He trained a CNN model on 3064 T1-weighted CE-MR images. He achieved 98.5% training accuracy and 84.19% validation accuracy(Abiwinanda et al. 2019).
Hossam et al. proposed a CNN architecture to classify different grades of brain tumours such as Grade II, Grade III and Grade IV. 3588 MR T1-weighted contrast-enhanced images were obtained. He achieved overall accuracies of 96.13% and 98.7% in two different studies(Sultan, Salem, and Al-Atabany 2019).Cinar and Yildirim et al. used pretrained models (ResNet50,AlexNet, DenseNet201,etc.) for the detection of brain tumors. Eight additional layers were added to ResNet50, replacing the last five layers. The total accuracy of this model was 97.2% (Çinar and Yildirim 2020). Khawaldeh et al. used the ConvNet model for classifying brain tumors. He used a modified version of AlexNet .He achieved 91.6% accuracy in the classification of brain tumors(Khawaldeh et al. 2017).
Talo et al. used a deep transfer learning approach for the classification of brain tumors. For this purpose he used the ResNet34 model. He used 613 MR images and achieved 100% 5- fold accuracy(Talo et al. 2019).Deepak and Ameer et al. used pretrained Google-Net models to extract features from M images of the brain. He obtained 98% accuracy in three different grades of brain tumors(Deepak and Ameer 2019). Mohsen et al. used a deep learning architecture for categorizing of four different grades of tumors. He used the DWT model for feature extraction and achieved 96.97% accuracy(Mohsen et al. 2018).
Deep learning has proven to be highly efficient in the field of biomedical sciences, surpassing traditional biomedical models by offering superior data descriptions(Rao, Sarabi, and Jaiswal 2015). The conventional method is more complex because the brain has high-density metabolites that are difficult to access manually. Consequently, many computer-based techniques are used to diagnose tumors (Reddy et al. 2020). These developments are vital resources for imaging experts as well as other medical specialties. MRI images are processed using a variety of algorithms based on deep learning, particularly for segmentation and image classification, which provides radiologists with valuable second opinions(Amin et al. 2018). The most popular deep learning method for brain tumor classification is convolutional neural network model (Nayak et al. 2022). Convolutional neural network models reduce the lengthy and time consuming manual image evaluation process. Most CNNs extract pertinent features from tumor images which leads to a substantial improvement in performance (Tripathy, Singh, and Ray 2023). The objective of this research is to effectively categorize tumors using Dense EfficientNet models which refers to CNNs models. This is the first time that EfficientNet models is used for classification of brain tumors into four different classes. EfficientNet models are more advanced network models and provide greater training accuracy on large datasets(Tan and Le 2019). Comprehensive image data are acquired to generate densely reconstructed segmentation masks for classifying tumors into four categories.