Abstract
Due to the complexity of the images and dearth of anatomical models, it is highly difficult to accurately represent the various deformations in each component of the medical images. In recent years, a significant number of children and adults have affected from brain tumors, which is one of the most terrible types of disease affects the people around the world. Moreover, the Magnetic Resonance Imaging (MRI) based brain tumor detection is one of a significant study area in the field of medical imaging. Since, the use of computerized methods aids in the detection and treatment of disease by the medical professionals. The development of an automated method for the accurate detection and classification of tumors from brain MRIs. In this framework, a tanh normalization process is used to smooth out the input brain MRIs with less noise artefacts and improved quality. Then, a group feature extraction model is used to extract the relevant features from the normalized image, which includes both Speeded Up Robust Features (SURF) and Grey Level Co-occurrence Matrix (GLCM) features. The Water Chaotic Fruitfly Optimization (WChFO) method is used to identify the best features for increasing the speed of classifier training and testing processes with less time. Moreover, a Deep Recurrent Neural Network (DRNN) model is used to classify the type of brain tumor for accurate early diagnosis and treatment. The most well-known benchmarking datasets, like BRATS and Kaggle, employed for analysis in order to assess the effectiveness and results of the proposed brain tumor diagnosis system. By using the proposed WChFO-DRNN technique, the accuracy of the tumor detection system is increased to 99.2% with the sensitivity, specificity of 99% and time consumption of 0.2s.
Introduction
Brain tumors, whether primary or metastatic (secondary), affect more than 190,000 people worldwide each and every year [1–3]. Although the exact etiology of brain tumor is unknown, there are numerous commonalities among those who develop them. Anyone can be impacted by it, whether they are a toddler or an adult. The risk of mortality has initially been found to be lower in the tumor region. As a result, the radiology division has become well-known for its imaging-based research on brain tumors [4, 5]. The causes of brain tumors have been the subject of numerous investigations, but the findings have not been clear-cut. Brain tumors are characterized by an abnormal cell proliferation in the human brain. Gliomas and meningiomas are two different types of tumors that can develop in the brain, spinal cord, or other body parts [6, 7]. Also, the Pituitary tumors are caused by the pituitary gland's aberrant cell proliferation. The identification and prognosis of brain tumors in many neurological disorders and conditions are made easier by the widely used MRI technology [8]. Based on visual characteristics and a contrast texture analysis of the soft tissue, standard MRI sequences are typically used to distinguish between various types of brain tumors. The World Health Organization (WHO) has classified more than 120 classes of brain tumors into four tiers based on their level of aggressiveness. According to histopathological characteristics, such as tissue, severity, and tumors rate of growth, the tumor [9, 10] is categorized into benign or malignant, usually the grade from low to high (1–4) are treated as malignancy. However, the use of basic MRI sequences may not be adequate for a correct diagnosis because gliomas are the most common primary brain tumors. The detection of brain cancers can be done more effectively with MRIs [11]. Due to manual processing, segmentation of MR images takes a long time in clinical settings. Also, the medical professionals must put a lot of effort and time for performing manual analysis [12, 13]. As a result, a fast and more accurate tumor diagnosis system requires the use of computerized techniques. Still, the development of an automated system is much more difficult, due to the variations in tumor thickness, appearance, and size. The tumor diagnosis model of standard CAD system is shown in Fig. 1.
Recently, some researchers have focused heavily on the interpretation of medical data using Artificial Intelligence (AI) mechanisms [14–16] such as machine learning and deep learning. The literature presents a number of completely automated and semi-automated methods for identifying and classifying brain tumors. Deep learning is a new machine learning technique that has gained a lot of interest in medical image processing, particularly for the diagnosis of brain tumors. Typically, deep learning methodology is used to divide the data into objective components and to assess each factor's accuracy. Only a limited number of investigations used different classifications from this useful resource to describe their findings. Deep learning [17] is a type of AI that controls calculations prompted by the capability and structure of the brain. However, the conventional ML & DL models facing the major challenges in high computational time, increased error predictions, overfitting, and low accuracy. Therefore, the proposed aims to implement a new detection framework for brain tumor diagnosis. The key contributions of the proposed work are as follows:
To smoothen the input brain MRIs with reduced noise artefacts and better quality, a tanh normalization operation is performed.
To extract the pertinent features from the normalized image, a group feature extraction model is deployed that holds both SURF and GLCM features.
To improve the speed of classifier training and testing with reduced time consumption, the optimal feature selection is performed with the use of Water Chaotic Fruitfly Optimization (WChFO) algorithm.
To categorize the class of brain tumor for early diagnosis and treatment with reduced error, a Deep Recurrent Neural Network (DRNN) model is implemented.
To assess the performance and outcomes of the proposed brain tumor diagnosis framework, the most popular benchmarking datasets such as BRATS and Kaggle are used for analysis.
The other portions of this paper are divided into the following categories: Section 2 presents the literature overview on the current machine learning and deep learning models for brain tumor detection and classification. Section 3 gives a description and stage-by-stage justifications for the proposed WChFO-DRNN system. In Section 4, the performance and comparison results are presented. In Section 5, the article is summarized with recommendations for further work.
Related works
The field of computer vision has recently given more attention to medical imaging. Brain tumor images are used by researchers to increase the segmentation and classification accuracy in medical imaging. They implemented an AI based automated systems for significantly performing brain tumor detection and categorization. The primary processes that are necessary for any automated system are the separation of the tumor from a given image, and the extraction of the necessary relevant information for classification. In this section, some of the recent works relevant to MRI brain tumor detection and classification are investigated with its pros and cons.
Arif et al. [18] implemented a new wavelet transformation based deep learning technique for the detection and segmentation of brain tumor. By employing the gray-level-co-occurrence matrix (GLCM) method, significant features are collected from each segmented tissue, and then a feature optimization process utilizing a genetic algorithm is performed. The suggested method for detecting tumors in MRI images works well. Here, the Berkeley wavelet transformation is used, which is particularly effective at identifying the area of interest in MR images because it employs a two-dimensional triadic wavelet transformation with a complete, arbitrary function. Rehman et al. [19] utilized a 3D-CNN model for developing the microscopic brain tumor detection framework. An important deep learning method for automatic feature extraction is CNN. Convolution, ReLu, Pooling, Normalization, and Fully Connection are only a few of the layers in a basic CNN model. Yet, significant resources, including a massive dataset and computational power, are needed to train a new model from the beginning. The suggested 3D CNN architecture is used to extract the tumor from MRI scans, then transfer learning is used to perform classification. Khan et al. [20] introduced an automated brain tumor detection system with the stages of tumor enhancement, segmentation, region extraction, feature extraction and classification. Typically, image segmentation is one of the crucial step in the field of emerging computer vision applications like medical imaging, video surveillance, and many others. However, segmentation of the abnormal region in medical imaging is even more crucial than any other. In the suggested framework, the marker based watershed segmentation approach is employed to segment the tumor affected portions with reduced segmentation error.
Khairandish et al. [21] introduced a hybridized model for improving the classification accuracy of brain tumor detection system. Here, the functions of standard CNN and SVM have been incorporated for brain tumor detection. In which, the CNN is specifically used to extract the features from the input MRIs, which is a kind of artificial neural network widely used in the medical imaging application systems. In this framework, the operations such as image resizing, skull removal, and noise filtering have been performed during preprocessing. Gurbina et al. [22] applied various wavelet transformation techniques for accurately spotting tumor from the brain MRIs. Due to their higher temporal resolution than Fourier transforms, they offer a significant benefit because they can record both intensity and position information in the images. Moreover, the Otsu thresholding model is applied for tumor segmentation, which is widely in the field of computer vision. Following a bimodal histogram, it determines the best threshold value to separate the image classes, so it highly maximizes the interclass variance and minimizes the mixed expansion. Furthermore, the SVM model is applied to categorize the brain image as healthy or tumor-affected. Banerjee et al. [23] intended to categorize both the High Grade Glioma (HGG) and Low Grade Glioma (LGG) for improving the process of diagnosis. For this purpose, the deep ConvNet model is implemented here, which trains the image from the beginning for accurate classification. The ConvNet architecture was created to sporadically resemble the basic operation of the human early visual system. It has been demonstrated that the visual cortex contains several abstraction layers that search for particular patterns in the sensory vision. A ConvNet is constructed using a similar concept of stacking many layers to enable it to learn various abstractions of the input data. However, it necessitates a significant amount of knowledge to choose the best model architecture and processing a large amount of training data, which could be the major limitation of this model.
Bhanothu et al. [24] utilized a deep convolutional network model for the detection and classification of brain tumor. This study used the Faster R-CNN algorithm to identify and categorize several types of brain tumors, where the VGG-16 base network architecture is used. This algorithm displays the confidence ratings together with the boundary for determining the tumor region. Rajinikanth et al. [25] introduced a Deep Learning Architecture (DLA) model for identifying brain abnormalities from MRIs. The classifier divides the considered images into the normal/tumor classes, which are mainly accountable for the DLA's overall performance. Chahal et al. [26] presented a comprehensive survey to examine several techniques used for brain tumor detection and classification. This review examines numerous segmentation/classification strategies that are effective for the early identification of a variety of brain disorders and attempts to assist researchers in determining the key characteristics of different forms of brain tumors. The manuscript discusses the most pertinent methodologies, methods, and associated preferences, limitations, and potential pitfalls for MR image-based brain tumor diagnosis. The current state of the art with regard to various tumor types could be summarized in an effort to aid researchers in determining new directions.
Because of its multiple layer architecture, which represents data with several levels of abstraction, it is able to address a number of issues that arise in conventional machine learning techniques. Due to its generalization and self-learning capabilities, the deep learning models enable better quantitative imaging feature analysis. Hence, it is more suitable for solving an improved neurological diagnosis problems. Deep learning-based classification approaches are therefore becoming more and more popular in the field of medical imaging. Siar et al. [27] presented a comprehensive analysis to examine various ML and DL models used for brain tumor diagnosis. Specifically, it described about the use of central clustering model for enhancing the performance of detection with better prediction outcomes. Maqsood et al. [28] implemented a fuzzy logic based U-Net segmentation model for developing an effective brain tumor diagnosis framework. Here, the dual tree complex WT mechanism is utilized for feature scaling, which helps to identify the edges in the brain images. Moreover, this framework includes the following operations:
Contrast enhancement
Wavelet transformation
Edge detection
Tumor classification
Segmentation
Sharif et al. [29] implemented an ELM model incorporated with the fuzzy set for categorizing the grade of tumor. Here, the image smoothening is performed with the use of triangular fuzzy median filtering technique, and the tumor affected region is segmented using the fuzzy based clustering model. Still, an automated brain tumor detection remains an ongoing research issue, and accuracy needs to be improved in order to identify brain tumors early and begin treatment. Moreover, the conventional model have the major problems correlated to the following factors: inaccurate detection rate, over segmentation, high error rate, high computational time, and error rate. Therefore, the proposed work motivates to develop a new automated disease diagnosis framework for brain tumor identification and class categorization.
Proposed methodology
This section provides the complete explanation for the proposed brain tumor diagnosis framework with the overall work flow and descriptions. The main contribution of this paper is to develop an automated and computationally efficient framework for brain tumor diagnosis. The work flow model of the proposed system is depicted in Fig. 2, where the separate image processing algorithms are applied to improve the overall tumor detection performance. The stages involved in this system are as follows:
Tanh normalization based preprocessing
Group Feature Extraction
Water Chaotic Fruitfly Optimization (WChFO)
Deep Recurrent Neural Network (DRNN)
Work flow of the proposed brain tumor detection framework
Citation: Imaging 15, 2; 10.1556/1647.2023.00122
Initially, the input brain MRIs are acquired from the public benchmarking datasets. Then, the preprocessing and tanh normalization operations are performed to filter the noise artefacts and to generate the smoothened brain image. After that, the group feature extraction model is applied to extract the possible and relevant features associated to the brain tumor, which includes the combination of SURF and GLCM features. In order to choose the best optimal features from the extracted feature vector, a novel and efficient WChFO algorithm is applied, which provides the optimal set of features for training and testing operations. Then, the Deep Recurrent Neural Network (DRNN) algorithm is applied to identify and categorize the grade of tumor according to the features.
Preprocessing
Feature extraction
In the domain of computer vision, feature extraction and reduction have proven crucial for classifying tumor region into appropriate categories. Finding the most relevant or robust features for classification that generate effective results is the main challenge in feature extraction. In the proposed framework, the different types of features such as SURF and GLCM are extracted from the normalized images for enhancing the accuracy of classification. Additionally, the filter outputs are normalized according to the mask size. This ensures that the Frobenius norm is consistent across all filter sizes. Image pyramids are frequently used to implement scale spaces. To reach the upper part of the pyramid, the images are periodically standardized with a Gaussian function. In the proposed work, the SURF and GLCM features are extracted from the preprocessed image, which are used for predicting the type of tumor, in which the SURF [30, 31] is one of the robust and fast model used for analyzing the similarity invariant representation of the MRI brain images. Moreover, it has the primary benefits of reduced computational time and increased accuracy, since it uses the hessian matrix for choosing the location and scale measures from the given images. During this matrix formation, the convolution of Gaussian second order derivative is estimated that are optimal for analysis. Scale-variable box filters and Gaussian kernel parameters are used to form the image pyramid. Each pixel ought to be contrasted with the remaining pixels and a pixel can be identified as an interest point if its value is higher than that of its neighbors. After constructing image pyramid, the interest points are localized for generating the descriptor. Consequently, the GLCM features [32, 33] are also extracted for obtaining the most useful features in order to maximize the accuracy of tumor detection. In this model, the distribution of measured intensities at specific locations in relation to one another in the image is used to compute the texture features. Statistics are divided into first-order, second-order, and higher-order statistics based on the quantity of pixels in every pair. Second order statistical texture features can be extracted using the GLCM approach. The technique has been employed in a variety of applications. Third and higher order textures incorporate the relationships between three or more pixels into consideration. Due to the complexity of the interpretation and the lengthy computation process, these are theoretically feasible but are rarely used. To determine the traits of texture statistics from the medical image, Haralick provides fourteen textural features based on the co-occurrence matrix. For instance, it includes the features of angular second moment, inverse difference moment, entropy, and correlation. In the proposed work, we are using both SURF and GLCM models for obtaining maximized tumor detection performance, hence it is referred to as group feature extraction model.
Water Chaotic Fruitfly Optimization (WChFO)
After image normalization, a computationally intelligent mechanism, named as, Water Chaotic Fruitfly Optimization (WChFO) algorithm is deployed for choosing the features for classifier training. It is a hybridization of Water-wave and Fruitfly optimization algorithms [34]. When compared to the other optimization algorithms, it specifically has the merits of high convergence, minimal time for fitness estimation, and high efficiency. Thus, the proposed work intends to apply the WChFO algorithm for improving the tumor classification accuracy of the proposed diagnosis framework. Typically, the fruitfly is a type of swarm intelligence optimization algorithm used for resolving the global optimization problems, which is developed based on the fruitfly's natural tendency to seek out food. This will utilize its osphresis capabilities to sense the smell of food. Fruit fly uses its abilities to detect smell to approach the food before moving closer and accessing it with its vision. Other flies may also move in the direction of food, depending on the flocking habits of the group. The optimal input weight values for the classifier's input weight parameter are selected using this characteristic for finding food. By choosing the ideal location, the fruitfly chaotic mapping improves the performance of the classification. On the other hand, water wave is a meta-heuristic model that illustrates water wave processes including shattering, dispersion, and propagation in order to derive the correct solution. It is a competitive meta-heuristic method that works well for locating the potential point. Moreover, the proposed WChFO algorithm comprises the following stages of operations:
Solution encoding
Parameter initialization
Position updation
Estimation of smell concentration
Chaotic operation
Identification of best solution
Tumor detection and classification
Sample images from BRATS dataset (a). Original image (b). Tumor detected region (c). Extracted tumor portion, and (d). Ground truth image
Citation: Imaging 15, 2; 10.1556/1647.2023.00122
Results and discussion
In this section, the performance and results of the proposed WChFO-DRNN based brain tumor diagnostic framework is validated and compared using several parameters. For this assessment, the most popular benchmarking MRI datasets such as BRATS and Kaggle have been used. There are 155 MRI samples with tumors among the total of 253 MRI samples in the Kaggle dataset [36], which has 845
According to image classification, the accuracy is a proportion that represents the total number of pixels that have been correctly classified in relation to the total number of pixels in the image. The precision indicates the likelihood of pixels being correctly categorized in addition to accuracy. Moreover, the other parameters such as sensitivity, specificity, f1-score, and dice coefficient are also considered as the most essential parameters used to assess the diagnosing performance of the classifiers. Table 1 and Fig. 8 presents the comparative analysis among the standard machine learning [18] and proposed classification models. Here, the error rate is also validated to determine that how accurately the classifier categorizes the labels with true positive rate and reduced false negative rate. The estimated results show that the combination of WChFO + DRNN technique provides an improved performance outcomes with reduced error rate.
Comparative analysis among the ML and proposed DL techniques
| Classifiers | Accuracy (%) | Error rate (%) | Sensitivity (%) | Specificity (%) | Precision (%) |
| NB | 90 | 10 | 86.6 | 84 | 93.3 |
| BoVW-SVM | 97.3 | 2.7 | 98.6 | 96.1 | 96 |
| CNN | 98.5 | 1.5 | 98.6 | 97 | 98 |
| Proposed | 99.2 | 0.9 | 99 | 99.1 | 99.4 |
Table 2 and Fig. 9 compares the detection accuracy of the existing [21] and proposed classification approaches using the BRATS dataset. Similarly, the accuracy is also estimated for the combination of existing [25] and proposed feature extraction based classification models as shown in Table 3. Since, feature extraction plays a major role in many medical imaging application systems, because the detection accuracy of classifier can be determined according to the number of features used for training and testing operations. In addition to that, the conventional ML and DL architecture [38] and proposed brain tumor detection models are validated and compared in terms of accuracy as shown in Table 4. Based on the overall obtained outcomes, it is evident that the tumor detection accuracy of the proposed WChFO-DRNN technique provides an increased accuracy, when contrasted to the other models.
Accuracy
| Detection Models | Accuracy (%) |
| RELM | 94.23 |
| DCNN | 95 |
| DNN-DWA | 96 |
| KNN | 96.6 |
| CNN | 97.5 |
| Proposed | 99.2 |
Accuracy analysis among ML and DL models
| Detection Models | Accuracy (%) |
| SVM + Fused Features | 97 |
| NB + Fused features | 98 |
| Ensemble + Fused features | 86 |
| DT + Fused features | 97 |
| KNN + Fused features | 89 |
| DL + Softmax | 97.8 |
| DL + MSVM | 93.6 |
| Shearlot Transform + PSO + SVM | 97.38 |
| CNN | 91.16 |
| Proposed | 99.2 |
Accuracy analysis among ML and DL models
| Detection Models | Accuracy (%) |
| CNN + GA | 94.20 |
| RELM | 94.23 |
| VGG 19 | 94.58 |
| GoogleNet | 97.10 |
| DBIRN | 99.69 |
| Proposed | 99.75 |
Table 5 and Fig. 10 presents the overall performance analysis of the proposed WChFO-DRNN based brain tumor diagnosis framework with respect to varying size of training data.
Performance analysis of the proposed model with respect to varying size of training data
| Training size (%) | Accuracy | Precision | Recall | F1-score |
| 50 | 99.7 | 99.69 | 99.51 | 99.50 |
| 60 | 99.72 | 99.69 | 99.53 | 99.51 |
| 75 | 99.75 | 99.7 | 99.59 | 99.55 |
Performance analysis of the proposed WChFO-DRNN model
Citation: Imaging 15, 2; 10.1556/1647.2023.00122
Moreover, Fig. 11 and Table 6 compares the standard machine learning [39] and proposed classification models using different parameters. Similarly, the accuracy, dice, sensitivity, and time measures of the brain tumor detection methodologies [40] are also validated and compared as shown in Table 7, and its graphical representation is shown in Figs 12 and 13. According to the overall observations and results, it is identified that the proposed WChFO-DRNN technique provides the better outcomes contrasted to the other methods. Compared to other studies in the analysis, the developed model performs remarkably well. Also, the suitable choice of preprocessing filters, optimization algorithms, the use of multi-scale feature extraction methods, and the development of an efficient DRNN classifier are all responsible for this excellent competency. Even the current classification methods give superior results for some input MR brain images, but they are unable to identify the tumor location in images with intricate structural properties. The sensitivity measures can be used to gauge how well tumor prediction by classification algorithms is performing. The proposed WChFO-DRNN model provides an accurate prediction of the tumor portion and other regions.
Comparative analysis
| Methods | Accuracy (%) | Precision (%) | Recall (%) | Specificity (%) |
| SVM | 96.8 | 97.5 | 95.12 | 97.43 |
| KNN | 93.3 | 93 | 93.46 | 93.03 |
| DT | 90.8 | 90.5 | 90.95 | 95.54 |
| Proposed | 99.12 | 99.35 | 99.38 | 99.51 |
Comparative analysis
| Methods | Accuracy | Dice | Sensitivity | Time (s) |
| RCNN | 92 | 87 | 95 | 0.47 |
| Faster RCNN | 94 | 94 | 94 | 0.25 |
| ResNet 50 | 95.9 | 95.5 | 95.3 | 0.20 |
| DenseNet 41 | 96.3 | 94.9 | 95.3 | 0.20 |
| Proposed | 99.1 | 99 | 99.1 | 0.15 |
Comparative analysis among the deep learning models
Citation: Imaging 15, 2; 10.1556/1647.2023.00122
The proposed brain tumor detection system comprises four stages of operations including preprocessing, feature extraction, feature selection, and classification. The preprocessing is initially performed after obtaining the input brain image, where the tanh normalization, feature weighting and noise removal operations are performed to smoothen the input image. Then, the group feature extraction model is deployed to extract the most required SURF and GLCM features from the normalized image. Consequently, the WChFO technique is applied to choose the most pertinent features from the extracted feature set, which optimally minimizes the dimensionality of image. Due to these operations, the training and testing complexity of the classifier is greatly reduced with minimal time consumption. Moreover, it helps to locate the tumor affected region with minimized false predictions, hence the proposed system obtains high accuracy up to 99.1% for the brain image dataset used in this study.
Conclusion
Today, digital image processing plays a crucial role in the healthcare system. The brain tumor diagnosis method is misleading because of human error when analyzing the numerous MRIs each day. To address this discrepancy and improve the precision of brain tumor treatment, computer-assisted methods are applied. For the precise detection of tumors, a variety of effective image preprocessing, segmentation, feature extraction and selection, and prediction techniques are crucial. This research develops an enhanced technique for dividing up the brain's tissue that requires little any human intervention. The major objective of this suggested approach is to let human experts or neurosurgeons quickly identify the tumor affected patients with reduced time. In the proposed framework, the public benchmarking datasets are initially used to obtain the input brain MRIs. To remove the noisy artefacts and produce the normalized brain image, preprocessing and tanh normalization operations are then carried out. Then, the group feature extraction model is used to extract any potential and pertinent characteristics related to the brain tumor. An innovative and effective WChFO method is used to choose the best features from the extracted feature vector, providing the best collection of features for training and testing activities. Finally the DRNN algorithm is then used to determine and classify the tumor grade based on its attributes. Moreover, a complete simulation analysis is carried out with the use of BRATS and Kaggle datasets to verify the efficacy and results of the proposed framework. Several evaluation measures have also been used to test the suggested model's performance. The findings obtained show that the WChFO-DRNN performs better than the recent state-of-the-art methods with reduced error and time.
In future, the current work can be developed by implementing a new deep learning based segmentation model for tumor diagnosis.
Authors' contribution: The author Rama Mohan Pasupuleti Contributed and put effort on Organize the Paper. Also technically contributed to data analysis. Sarvade Pedda Venkata Subba Rao technically contributed and made English Corrections and grammar checking. The author Prema Kothandan involved in the Background study of the Paper and helped the mathematical derivations. The author Samarthi Swapna Rani technically involved and provided a factual review and helped edit the manuscript. The author Sathees Babu Shanmuganathan technically involved and helped to derive the mathematical equation. The author Vijayaprabhu Arumugam contributed over all manuscript Review and Revision of the Manuscript.
Conflict of interest: All authors there is no Conflict of Interest to Publish this article.
Funding sources: The authors have no relevant financial or non-financial interests to disclose.
Ethical statement: The study submitted to Imaging has been conducted in accordance with the Declaration of Helsinki and according to requirements of all applicable local and international standards.
Acknowledgements
I am grateful to all of those with whom I have had the pleasure to work during this and other related Research Work. Each of the members of my Dissertation Committee has provided me extensive personal and professional guidance and taught me a great deal about both scientific research and life in general.
References
- [1]↑
Tiwari A, Srivastava S, Pant M: Brain tumor segmentation and classification from magnetic resonance images: review of selected methods from 2014 to 2019. Pattern Recognit Lett 2020; 131: 244–260.
- [2]
Nazir M, Shakil S, Khurshid K: Role of deep learning in brain tumor detection and classification (2015 to 2020): a review. Comput Med Imaging Graph 2021; 91: 101940.
- [3]
Harmouche A, Kövér F, Szukits S, Dóczi T, Bogner P, Tóth A: WebMRI: brain extraction and linear registration in the web browser. Imaging 2023. https://doi.org/10.1556/1647.2023.00111.
- [4]↑
Narmatha C, Eljack SM, Tuka AARM, Manimurugan S, Mustafa M: A hybrid fuzzy brain-storm optimization algorithm for the classification of brain tumor MRI images. J Ambient Intell Humaniz Comput 2020: 1–9.
- [5]↑
Khan HA, Jue W, Mushtaq M, Mushtaq MU: Brain tumor classification in MRI image using convolutional neural network. Math Biosci Eng 2020; 17: 6203–6216.
- [6]↑
Badža MM, Barjaktarović MČ: Classification of brain tumors from MRI images using a convolutional neural network. Appl Sci 2020; 10; 1999.
- [7]↑
Tandel GS, Balestrieri A, Jujaray T, Khanna NN, Saba L, Suri JS: Multiclass magnetic resonance imaging brain tumor classification using artificial intelligence paradigm. Comput Biol Med 2020; 122: 103804.
- [8]↑
Arbane M, Benlamri R, Brik Y, Djerioui M: Transfer learning for automatic brain tumor classification using MRI images. In: 2020 2nd international workshop on human-centric smart environments for health and well-being. IHSH, 2021, pp. 210–214.
- [9]↑
Alanazi MF, Ali MU, Hussain SJ, Zafar A, Mohatram M, Irfan M, et al.: Brain tumor/mass classification framework using magnetic-resonance-imaging-based isolated and developed transfer deep-learning model. Sensors 2022; 22: 372.
- [10]↑
Jia Z, Chen D: Brain tumor identification and classification of MRI images using deep learning techniques. IEEE Access 2020.
- [11]↑
Wickramasinghe SU, Weerakoon TI, Gamage PJ, Bandara MS, Pallewatte A: Identification of radiomic features as an imaging marker to differentiate benign and malignant breast masses based on magnetic resonance imaging. Imaging 2022.
- [12]↑
Bezdan T, Zivkovic M, Tuba E, Strumberger I, Bacanin N, Tuba M: Glioma brain tumor grade classification from mri using convolutional neural networks designed by modified fa. In: Intelligent and Fuzzy Techniques: Smart and Innovative Solutions: Proceedings of the INFUS 2020 Conference, Istanbul, Turkey, July 21-23, 2020, 2021, pp. 955–963.
- [13]↑
Chelghoum R, Ikhlef A, Hameurlaine A, Jacquir S: Transfer learning using convolutional neural network architectures for brain tumor classification from MRI images. In: Artificial Intelligence Applications and Innovations: 16th IFIP WG 12.5 International Conference, AIAI 2020, Neos Marmaras, Greece, June 5–7, 2020, Proceedings, Part I 16, 2020, pp. 189–200.
- [14]↑
Kumar S, Mankame DP: Optimization driven deep convolution neural network for brain tumor classification. Biocybern Biomed Eng 2020; 40: 1190–1204.
- [15]
Saleh A, Sukaik R, Abu-Naser SS: Brain tumor classification using deep learning. In: 2020 International Conference on Assistive and Rehabilitation Technologies (iCareTech), 2020, pp. 131–136.
- [16]
Mehrotra R, Ansari M, Agrawal R, Anand R: A transfer learning approach for AI-based classification of brain tumors. Mach Learn Appl 2020; 2: 100003.
- [17]↑
Chitra T, Sundar C, Gopalakrishnan S: Investigation and classification of chronic wound tissue images using random forest algorithm (RF). Int J Nonlinear Anal Appl 2022; 13: 643–651.
- [18]↑
Arif M, Ajesh F, Shamsudheen S, Geman O, Izdrui D, Vicoveanu D: Brain tumor detection and classification by MRI using biologically inspired orthogonal wavelet transform and deep learning techniques. J Healthc Eng 2022: 2022.
- [19]↑
Rehman A, Khan MA, Saba T, Mehmood Z, Tariq U, Ayesha N: Microscopic brain tumor detection and classification using 3D CNN and feature selection architecture. Microsc Res Tech 2021; 84: 133–149.
- [20]↑
Khan MA, Lali IU, Rehman A, Ishaq M, Sharif M, Saba T, et al.: Brain tumor detection and classification: a framework of marker-based watershed algorithm and multilevel priority features selection. Microsc Res Tech 2019; 82: 909–922.
- [21]↑
Khairandish MO, Sharma M, Jain V, Chatterjee JM, Jhanjhi N: A hybrid CNN-SVM threshold segmentation approach for tumor detection and classification of MRI brain images. Irbm 2022; 43: 290–299.
- [22]↑
Gurbină M, Lascu M, Lascu D: Tumor detection and classification of MRI brain image using different wavelet transforms and support vector machines. In: 2019 42nd International Conference on Telecommunications and Signal Processing (TSP), 2019, pp. 505–508.
- [23]↑
Banerjee S, Mitra S, Masulli F, Rovetta S: Deep radiomics for brain tumor detection and classification from multi-sequence MRI. arXiv preprint arXiv:1903.09240, 2019.
- [24]↑
Bhanothu Y, Kamalakannan A, Rajamanickam G: Detection and classification of brain tumor in MRI images using deep convolutional network. In: 2020 6th international conference on advanced computing and communication systems (ICACCS), 2020, pp. 248–252.
- [25]↑
Rajinikanth V, Joseph Raj AN, Thanaraj KP, Naik GR: A customized VGG19 network with concatenation of deep and handcrafted features for brain tumor detection. Appl Sci 2020; 10: 3429.
- [26]↑
Chahal PK, Pandey S, Goel S: A survey on brain tumor detection techniques for MR images. Multimed Tools Appl 2020; 79: 21771–21814.
- [27]↑
Siar M, Teshnehlab M: Brain tumor detection using deep neural network and machine learning algorithm. In: 2019 9th international conference on computer and knowledge engineering (ICCKE), 2019, pp. 363–368.
- [28]↑
Maqsood S, Damasevicius R, Shah FM: An efficient approach for the detection of brain tumor using fuzzy logic and U-NET CNN classification. In: Computational Science and Its Applications–ICCSA 2021: 21st International Conference, Cagliari, Italy, September 13–16, 2021, Proceedings, Part V 21, 2021, pp. 105–118.
- [29]↑
Sharif M, Amin J, Raza M, Anjum MA, Afzal H, Shad SA: Brain tumor detection based on extreme learning. Neural Comput Appl 2020; 32: 15975–15987.
- [30]↑
Purwar RK, Srivastava V: A novel feature based indexing algorithm for brain tumor MR-images. Int J Inf Technol 2020; 12: 1005–1011.
- [31]↑
Bhagat P, Choudhary P, Singh KM: A comparative study for brain tumor detection in MRI images using texture features. In: Sensors for health monitoring. Elsevier, 2019, pp. 259–287.
- [32]↑
Hussain A, Khunteta A: Semantic segmentation of brain tumor from MRI images and SVM classification using GLCM features. In: 2020 second international conference on inventive research in computing applications (ICIRCA), 2020, pp. 38–43.
- [33]↑
Sooch SK, Kapoor N: Brain Tumor detection with GLCM feature extraction and hybrid classification approach. In: Soft computing: theories and applications: proceedings of SoCTA 2022. Springer, 2023, pp. 37–46.
- [34]↑
Ingaleshwar S, Dharwadkar NV: Water chaotic fruit fly optimization-based deep convolutional neural network for image watermarking using wavelet transform. Multimed Tools Appl 2021: 1–25.
- [35]↑
Zhou Y, Li Z, Zhu H, Chen C, Gao M, Xu K, et al.: Holistic brain tumor screening and classification based on densenet and recurrent neural network. In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries: 4th International Workshop, BrainLes 2018, Held in Conjunction with MICCAI 2018, Granada, Spain, September 16, 2018, Revised Selected Papers, Part I 4, 2019, pp. 208–217.
- [36]↑
Tamilselvi R, Nagaraj A, Beham MP, Sandhiya MB: Bramsit: a database for brain tumor diagnosis and detection. In: 2020 Sixth International Conference on Bio Signals, Images, and Instrumentation (ICBSII), 2020, pp. 1–5.
- [37]↑
Weninger L, Rippel O, Koppers S, Merhof D: Segmentation of brain tumors and patient survival prediction: methods for the brats 2018 challenge. In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries: 4th International Workshop, BrainLes 2018, Held in Conjunction with MICCAI 2018, Granada, Spain, September 16, 2018, Revised Selected Papers, Part II 4, 2019, pp. 3–12.
- [38]↑
Kokkalla S, Kakarla J, Venkateswarlu IB, Singh M: Three-class brain tumor classification using deep dense inception residual network. Soft Computing 2021; 25: 8721–8729. 2021/07/01.
- [39]↑
Nanmaran R, Srimathi S, Yamuna G, Thanigaivel S, Vickram A, Priya A, et al.: Investigating the role of image fusion in brain tumor classification models based on machine learning algorithm for personalized medicine. Comput Math Methods Med 2022: 2022.
- [40]↑
Masood M, Nazir T, Nawaz M, Mehmood A, Rashid J, Kwon H.-Y, et al.: A novel deep learning method for recognition and classification of brain tumors from MRI images. Diagnostics 2021; 11: 744.
