The Evaluation of 2D and EfficientB0 Convolution Networks for detecting Brain tumor based on MRI images.

. Brain tumors represent a significant healthcare challenge, affecting both children and adults with potentially aggressive consequences. Accounting for a substantial percentage of all primary Central Nervous System (CNS) tumors, brain tumors pose a substantial burden, with approximately 11,700 new diagnoses annually. The classification of brain tumors into benign, malignant, pituitary, and other types necessitates precise diagnostic techniques and treatment planning to enhance patient life expectancy. Traditionally, the detection of brain tumors relied on the expertise of specialists analyzing Magnetic Resonance Images (MRI) without the aid of advanced technology. MRI remains the gold standard for brain tumor detection, generating vast amounts of image data for radiologists to interpret. Manual examinations, however, carry a risk of errors due to the intricacies and diverse properties of brain tumors, potentially leading to delayed treatment and, tragically, loss of lives. In this context, the application of automated classification techniques using Machine Learning (ML) and Deep Learning (DL) has emerged as a promising solution. These techniques, primarily employing Deep Learning Algorithms such as Convolutional Neural Networks (CNN) like 2D-convolutions and Deep Learning Models like ResNet50 and EfficientNetB0, in addition to traditional Machine Learning algorithms like Support Vector Machines (SVM), have consistently demonstrated superior accuracy in brain tumor detection compared to manual prediction. These automated methods have consistently exhibited superior accuracy in the detection and classification of brain tumors compared to manual approaches. This research proposes a robust system for the early detection and accurate classification of brain tumors, leveraging the power of Deep Learning and Machine Learning. By incorporating state-of-the-art


Introduction
The World Health Organization estimates that approximately less than 1% of the population will be diagnosed with this sort of tumor in the course of their lives.85% to 90% of primary cancers of the central nervous system (CNS) are brain tumors [1].2020 witnessed an anticipated 308,102 cases of primary brain or spinal cord tumors identified worldwide.A brain tumor is an irregular mass or abnormal growth of cells that develops within the brain or the central nervous system (CNS).These tumors can be categorized as either benign, meaning non-cancerous, or malignant, indicating a cancerous nature.Brain tumors may originate from various cell types within the brain and can manifest in different regions of the CNS, encompassing the brain itself, the spinal cord, or the meninges, which are the protective membranes encasing the brain and spinal cord [2].
The causes of brain tumors are multifaceted and, in many cases, not fully understood.While the exact origins of these tumors can be complex, several potential factors have been associated with their development.Exposure to ionizing radiation, such as high-dose radiation therapy, environmental radiation, or occupational exposure, is a known risk factor for certain types of brain tumors.Genetic predisposition and hereditary conditions can also contribute to an increased risk of developing brain tumors in some individuals.Furthermore, certain rare genetic syndromes are associated with a higher incidence of brain tumors [3].
Although these factors are recognized, it's crucial to note that many cases of brain tumors have no clear or identifiable cause, and the precise mechanisms leading to tumor formation remain a subject of ongoing research.Additionally, the interplay of genetic, environmental, and lifestyle factors in brain tumor development is a complex area of study that continues to be explored by researchers and medical professionals.Early detection, appropriate medical care, and a comprehensive understanding of these contributing factors are essential in addressing this health challenge effectively.Modern technology is essential for the early identification and detection of brain cancers.Modern neuroimaging methods, such Computed Tomography (CT) scans and Magnetic Resonance Imaging (MRI), provide extensive details on the internal structure of the brain and enable healthcare professionals to precisely identify the location, size, and features of tumors.Medical professionals can examine data more precisely when they use computer-aided diagnostic tools and advanced imaging software.The utilization of artificial intelligence (AI) and machine learning algorithms is extremely valuable in detecting microscopic inconsistencies that could be difficult for human observers to notice.

Related Works
Radiologists are using artificial intelligence in various areas of brain tumor research.Numerous techniques, including transfer learning and deep learning, are used to diagnose brain tumors.CNN architecture is used for segmentation and classification.These techniques offer a significant lot of understanding regarding diagnosis, treatment, and perception in radiology.Through the use of transfer learning-based classification, this study aims to address the current segmentation issues and produce high-quality results with reduced computation time and error rate without the need for specialized hardware, which is unavailable in developing nations with multiple image processing tasks equipped for MRIs with focal disabilities.Preprocessing and post processing are carried out for improved outcomes of the suggested model.The GoogleNet model is trained using transfer learning techniques to classify the MR images [4].
One of the most important tasks that radiologists or clinical professionals perform daily is the identification, segmentation, and extraction of contaminated tumor regions from Magnetic Resonance Imaging (MRI) pictures.Different human organ anatomical structures can be imagined using image-processing concepts.Basic imaging techniques have difficulty detecting aberrant structures in the human brain.In this work, a fully automatic heterogeneous segmentation method utilizing support vector machines (FAHS-SVM) based on deep learning approaches is suggested for brain tumor segmentation.The current study suggests using MRI imaging to separate the entire cerebral venous system and add a new, totally automatic method based on anatomical, morphological, and relaxometry details.A notable characteristic of the segmenting function is the high degree of homogeneity between the surrounding brain tissue and the architecture.One or more layers of hidden nodes make up the ELM learning algorithm type.Regression and classification are two domains in which these networks find use.The probabilistic neural network classification method has been used to train and validate the tumor detection accuracy in brain MRI images.The numerical results indicate that the brain's aberrant and normal tissues may be distinguished with around 98.51% accuracy.Pictures from magnetic resonance show how effective the technology is [5][6][7].
The goal of this research is to create an automated tumor classification model that will help radiologists diagnose brain tumors more accurately.Because a timely tumor diagnosis is often impossible to achieve, brain tumors are fast rising to the top of the global cause of death list.By automatically extracting and classifying features, the proposed Caps-VGGNet hybrid model combines the capabilities of the CapsNet and VGGNet models to handle the difficulty of massive dataset requirements.Based on CE-MRI images from the Brats-2020 and Brats-2019 datasets, the research presents a novel hybrid method that makes use of deep learning approaches to identify, extract radiomic characteristics, and classify brain cancer.Preprocessing procedures were used to increase the quality of the photos, including distant image improvement techniques.In order to extract complicated information from the images and enable early detection and precise classification of brain cancer, transfer learning-based architectures were optimized.By utilizing a hybrid CNN-based architecture and SoftMax classifier layers, the suggested method outperforms the state-of-the-art techniques in terms of true positive rate and false positive rate [8][9][10][11][12][13][14][15].
In this study, the classification of tumor-and normal-brain tissue is done using ANN and CNN.Artificial Neural Networks, or ANNs, operate similarly to the nervous system of the human brain.This is achieved by connecting many digital computers through networking and interconnections, which enables neural networks to be trained using simple processing units applied to training data and store experiential knowledge.Algorithms from machine learning and deep learning are used to detect brain tumors.When these algorithms are used on MRI pictures, brain tumors can be predicted quickly, and better accuracy helps patients receive treatment.The radiologist can also make quick decisions with the aid of these predictions.The suggested work applies a self-defined Artificial Neural Network (ANN) and a convolution neural network (CNN) to detect brain tumors and analyses their performance [16][17][18][19].
Early brain tumor detection is a critical task for radiologists.The average size of a brain tumor doubles in just twenty-five days.If not treated properly, the patient's survival rate is usually less than half a year.It can quickly result in death.As a result, an automatic system is required for early detection of brain tumors.In this paper, an automated method for distinguishing between cancerous and non-cancerous Magnetic Resonance Imaging (MRI) of the brain is proposed.To identify and categories brain tumors at the picture and lesion levels, an automatic approach is put forth [20][21][22][23][24][25].The suggested methodology is divided into multiple parts, such as feature extraction, classification, and preprocessing.Various techniques are used in the preprocessing step to segment the candidate lesions.Then, for each candidate lesion, a set of characteristics based on texture, shape, and intensity is applied.Together with shape and intensity parameters, texture features are useful descriptors of the tumor because a lesion's combination of texture, shape, and intensity provides superior discrimination information.To select a better classification strategy, several tests are run on the chosen feature set.Support vector machines (SVM) with a geometrical family are used for this, using various crossvalidations [26][27][28][29].

Problem Statement:
The healthcare sector is totally different from other industries.It is a high priority sector and people expect the highest level of care and services regardless of cost.The brain is an organ that controls activities of all the parts of the body.After the success of deep learning in other real-world applications, it is also providing exciting solutions with good accuracy for medical imaging and is a key method for future applications in health sector.The brain is an organ that controls activities of all the parts of the body.Recognition of automated brain tumor in Magnetic resonance imaging (MRI) is a difficult task due to complexity of size and location variability.

Objective:
In order to help the medical professionals, we propose a model that detects a tumor from a MRI (Magnetic Resonance Imaging) scan.As we know humans make errors, can't detect.tumor at early stages and the error in analyzing scan can cost a person's life.The model is trained with a training dataset containing the MRI (Magnetic Resonance Imaging) images, the target variable being the MRI images has tumor or not in it.Various deep learning algorithms like CNN-2D and deep learning models like efficientnetB0, ResNet50 are used for training the model and then the model is tested using the testing dataset, the algorithms or model which gives least error and more accuracy is considered.This system aims to make a meaningful impact in the field of healthcare, improving the lives of patients and facilitating effective treatment.

System Architecture:
System designers and developers often utilize a basic architecture diagram, typically created using Unified Modeling Language (UML), to visually represent the overall structure of a system or application.This diagram serves multiple purposes, including verifying that the system meets user requirements and describing design patterns within the system.The architecture diagram allows individuals to abstract the core framework of a software system, establishing boundaries, relationships, and constraints among its components.By providing a comprehensive view of the system's physical deployment and future development plans, it aids in understanding the system's overall design.Developers and designers find this diagram highly valuable as it offers a clear and concise representation of the system's architecture.It helps them communicate and collaborate effectively, ensuring that the system is well- designed and aligned with user needs.Figure 10 shows the architecture diagram of the system which consists of all the parts that are performed to construct the system.Here we have a MRI image dataset then these images undergo preprocessing, build the models with the required algorithms and train the models with the dataset and at last test the model for accuracy.

Efficientnetb0:
EfficientNetB0 is a convolutional neural network architecture that aims to achieve higher accuracy and efficiency by scaling the model's depth, width, and resolution in a systematic manner.Developed to address the trade-off between model size and performance, EfficientNetB0 employs a novel compound scaling method that uniformly scales all dimensions of the network, including depth, width, and resolution, to optimize the model's overall efficiency.By strategically balancing these dimensions, the model can achieve better accuracy while maintaining a manageable model size, making it well-suited for resourceconstrained environments.EfficientNetB0 has demonstrated remarkable performance across various computer vision tasks, including image classification and object detection, outperforming other existing models with similar computational costs.Its scalability and robust performance have established EfficientNetB0 as a prominent architecture in the field of deep learning, serving as a benchmark for developing efficient and accurate convolutional neural networks for image-related applications [9].EfficientNetB0, a member of the EfficientNet family of models, is based on compound scaling that efficiently balances the model's depth, width, and resolution.While the architecture involves intricate design choices and compound scaling coefficients, the mathematical equation at the core of its architecture can be represented in a simplified form.Let D, W, and R denote the depth, width, and resolution scaling coefficients, respectively.Considering an input image I and the model's learnable parameters effWeff, the output of EfficientNetB0 can be represented as: underlying mechanism through which the compound scaling coefficients D, W, and R influence the architecture, enabling EfficientNetB0 to strike a balance between model complexity, computational efficiency, and overall performance.By carefully adjusting these scaling parameters, EfficientNetB0 can achieve optimal accuracy and efficiency for various image-related tasks.
In figure EfficientNet model architecture is shown which contains several layers.

Resnet50:
ResNet50, short for Residual Network with 50 layers, is a variant of the deep convolutional neural network architecture that has made significant strides in the field of computer vision.It addresses the challenge of training very deep neural networks by introducing residual connections, which allow the network to learn residual mappings instead of directly trying to learn unreferenced mappings [10].This architecture consists of multiple convolutional layers with shortcut connections that enable the smooth flow of information throughout the network, preventing the vanishing gradient problem and facilitating the training of extremely deep models.ResNet50 has demonstrated exceptional performance in tasks such as image classification, object detection, and image segmentation, owing to its ability to extract intricate features from complex datasets.Its design has significantly contributed to the advancement of deep learning in computer vision and has become a standard benchmark for various image-related applications, showcasing its efficacy and robustness in handling challenging visual recognition tasks [11].The ResNet50 architecture involves the use of residual blocks, which can be mathematically represented as follows: Given an input feature map X and the weights of the residual block, the output of the block Y can be expressed as: Y=F(X,{Wi})+X where F denotes the residual function, Wi represents the weights within the residual block, and ++ denotes the element-wise addition.The residual function allows the network to learn the residual mapping, which helps in addressing the vanishing gradient problem and facilitates the training of deeper networks.This mechanism enables the network to learn more intricate representations by building on the existing features, promoting the effective training of deep neural networks.ResNet50, with its 50 layers, incorporates multiple residual blocks to create a powerful and effective deep learning architecture for tasks such as image classification and object recognition.

CNN 2D Algorithm:
The 2D convolution algorithm serves as a vital operation in the realm of image processing and computer vision.It involves the application of a 2D filter or kernel to an input image, facilitating the extraction of specific features or the execution of image manipulation tasks like blurring, sharpening, or edge detection.This process necessitates the systematic sliding of the kernel over the image, performing element-wise multiplications between the corresponding pixel values and the filter elements at each position, and subsequently summing up the outcomes to derive the value of the output pixel.Through this iterative procedure encompassing every pixel in the image, a new output image is generated, emphasizing distinct image attributes or characteristics based on the inherent traits of the specific filter in use [12].With its crucial role in tasks such as feature extraction, image enhancement, and object detection within computer vision, the 2D convolution algorithm functions as a foundational component, forming the basis for more intricate deep learning architectures involved in activities like image segmentation and classification.The mathematical equation for 2D convolution can be expressed as follows.
Given an input 2D signal or image I of size m×n and a 2D filter or kernel K of size k×l, the resulting 2D convolution output S can be computed as: where S represents the resulting output image, I denotes the input image, K is the filter, and u and v are the indices used for iterating over the dimensions of the filter.The indices i and j iterate over the spatial dimensions of the input image, with 0≤i<m−k+1and 0≤j<n−l+1, considering the boundaries of the image and the filter.This equation essentially performs an element-wise multiplication between the corresponding elements of the filter and the image, followed by a summation of the results over the appropriate ranges, producing the value for each pixel in the output image.

CNN:
For the analysis and processing of visual data, including images and videos, the Convolutional Neural Network (CNN) algorithm is a powerful deep learning technique.Convolutional, pooling, and fully connected layers are among the separate layers that make up this structure.Each layer has a specific purpose in identifying and acquiring complex hierarchical representations of the input data.Convolutional layers, which apply various filters or kernels to the input image to aid in the identification of particular features and patterns, are a key component of CNNs [13].CNNs can reduce the spatial dimensions of the data and extract important details with minimal computational complexity by utilising pooling layers and activation functions.To create predictions or classifications, the fully connected layers at the end of the network use the features that have been extracted.CNNs are an essential component of computer vision and image analysis because of their exceptional performance in tasks like object detection, image segmentation, and image classification [14].
The mathematical equation for a Convolutional Neural Network (CNN) involves several key components, including the convolution operation, the activation function, and the pooling operation.Given an input image I and a set of learnable filters or kernels W, the convolution operation in a CNN can be mathematically represented as.where C denotes the output of the convolutional layer, f represents the activation function (such as ReLU or Sigmoid), i and j are the spatial indices of the output feature map, and m and n iterate over the spatial dimensions of the filter.The term b denotes the bias term associated with the convolutional layer.In addition to the convolutional operation, CNNs often incorporate pooling layers to reduce the spatial dimensions of the feature maps.The mathematical equation for the pooling operation can be expressed as follows.

P(i,j)=PoolingFunction({C(i′,j′):i′∈[i,i+p),j′∈[j,j+q)})
where P represents the output of the pooling layer, PoolingFunction denotes the pooling function (such as max pooling or average pooling), and p and q are the dimensions of the pooling window.These mathematical equations illustrate the fundamental operations involved in a CNN, demonstrating how the network processes and transforms input data to extract meaningful features.Below figure show the internal functioning of 2 dimensional covolution neural network.The above table contains the confusion matrix of the three algorithms which are used for tumor detection.
The above table shows the validation and training accuracy graphs, the graph is plotted for epochs the model is trained and validated.The above table consists of the classification report s of all the algorithms used to build the system for tumor tumor detection.The table shows the accuracy of the algorithms and models used for tumor detection and out of all EfficientNet give highest accuracy.

Table 5. Performance metrics
The above table consists of all the performance metrics of the algorithms used to build the system for brain tumor detection.

Conclusion
This research proposes a groundbreaking initiative in the realm of brain tumor detection and classification.By leveraging the capabilities of Deep Learning and Machine Learning, our system aims to redefine the diagnostic process, moving beyond the limitations of manual interpretation.With the integration of advanced algorithms, including Convolutional Neural Networks (CNN), ResNet50, EfficientNetB0, and traditional methodologies like Support Vector Machines (SVM), our goal is to establish a reliable and precise mechanism for early identification of brain tumors.This shift towards automation seeks to address the potential errors associated with manual examinations, leading to quicker and more effective treatment interventions.The impact of this innovative approach extends to the global healthcare landscape, offering a powerful tool for medical professionals to enhance the accuracy and efficiency of brain tumor diagnosis.Ultimately, the integration of cutting-edge technologies holds the promise of transforming patient outcomes by facilitating timely interventions and alleviating the challenges posed by brain tumors in healthcare.

Fig. 3 .
Fig. 3. Efficient Net where O represents the output of the EfficientNetB0 model.This equation symbolizes the

Fig. 4 .
Fig. 4. ResNet Above figure shows resnet model architecture which consists many layers

Table 2 .
Training and Validation accuracy Graph

Table 3 :
Accuracy of algorithms