Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 330 https://internationalpubls.com Reducing loss for Brain tumour detection and classification in MRI using deep learning techniques Venkata Ramana Saddi1, M. Sunil Kumar2 1Technology Lead, ACE American Insurance Company CHUBB, Raleigh, NC, USA, ramana.saddi@gmail.com ramana.saddi@outlook.com 2Professor & Programme Head, Department of CSE,School of Computing, Mohan Babu University (erstwhile Sree Vidyanikethan Engineering College), Tirupathi, AP, India. sunilmalchi1@gmail.com Article History: Received: 11-06-2024 Revised: 10-07-2024 Accepted: 31-07-2024 Abstract: The signs of a brain tumour might be general or specific. The typical symptom is brought on by a tumour pressing against the brain or spinal cord. When the tumour has impacted a certain section of the brain and that area is not working correctly. Identification of brain tumours was extremely difficult and complicated due to the location, kind, shape, and size of the tumour in the brain. Brain tumour diagnosis is difficult since the tumor's size and resolution cannot be clearly evaluated in the early stages of growth. However, if a tumour is found and diagnosed early in the stage of tumour growth, there is a good chance that the patient will be treated. As a result, early tumour diagnosis is essential for successful treatment. To solve this problem, we suggested an automated method that can more help detect and classify three different types of brain tumours on MRI using deep learning techniques, faster RCNN+ResNet50. Keywords: Brain tumour, detection, classification, deep learning, faster RCNN, ResNet- 50, automated, MRI. 1. Introduction Due to its rising presence and high death rate across all age groups [1] [2] [3][4][5], brain tumours are among the world's worst illnesses [6] [7]. It is noted in [3] as India second-most frequent type of cancer. 5 to 10 brain tumours per 100,000 persons are seen in India. According to "International Association of Cancer Registries (IARC)," over 28,000 instances of brain tumours are detected more than 24,000 individuals die due to brain tumour every year in India. [8]. Increased usage of electronic devices like mobile phones, tablets, and other devices, has made this disorder more prevalent among children as well[9]. 1.1 Brain tumour Two kinds of brain tumours exist: benign (noncancerous) and malignant (cancerous). The patient's health may deteriorate as a result of the malignant tumours' fast spread to other brain areas [10]. When the majority of the cells are worn out or harmed, they are destroyed and changed out with fresh ones. Problems may arise if damaged and old cells are not removed when producing new ones. The growth or tumour is referred to as a mass of tissue that forms as extra cells are produced. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 331 https://internationalpubls.com 1.2 Brain tumour causes Although cause of the brain tumour is frequently unclear, table 1 specifies the things may increase a person's chance of getting one [17]. Table 1: Risk factors which cause brain tumour Risk Factors Supportive reasons Head injury and seizures There has long been research on the relationship between brain tumours and serious head injuries. Ionizing radiation It has been demonstrated that prior ionising radiation therapy, including x- rays, to the head or brain increases the likelihood to develop brain tumour. Electromagnetic fields Majority of research examining the effect of electromagnetic fields, like energy generated via power lines or by using a cell phone, don't find any evidence of a connection to an elevated risk of adult brain tumour development. Exposure to allergen, viruses and infections In tissue of brain tumours, high concentrations of the common virus cytomegalovirus (CMV) have been discovered. Family history Brain tumours may be caused by inherited genetic factors or circumstances in around 5% of cases. Home and work exposures Exposure to lubricants, pesticides, oil-based chemicals, latex, or vinyl chloride increases the risk of brain tumour. Gender In general, men are more likely then women to get the brain tumour. One specific type of brain tumour which affects women most frequently is the meningioma. Age Although anybody can acquire a brain tumour, youngsters and older individuals are more likely to do so. 1.3 Brain tumour symptoms and signs Brain tumour symptoms might be generic or specialized. The common symptom is caused by tumor's pressure on spinal cord or brain. When the particular area of brain is affected by the tumour and is not functioning properly, distinct symptoms are brought on [18]. General symptoms includes: • Pressure/head ache near the tumour • Partial or complete vision loss • A change in one's capacity for walking or carrying out everyday tasks • Sleep problems • Memory problems • Loss of balance • Difficulty in swallowing Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 332 https://internationalpubls.com • Fatigue • Vomiting • Seizures 1.4 Brain tumour detection Due to location, kind of tumour, form and size in brain, identification of brain tumours was exceedingly intricate and challenging. Since the tumor's size and resolution cannot be precisely measured in the early stages of development, diagnosing brain tumours is challenging [11]. However, the likelihood of a patient being treated is quite great if tumour is identified and diagnosis early in tumour growth phase. As the result, early tumour detection is crucial for effective therapy [12]. The diagnosis is frequently made by the physical examination utilizing magnetic imaging or computer tomography (CT). Most popular, crucial techniques to identify and assess patient brain is MRI imaging, which produces precise pictures of the brain. MRI images outperform other imaging modalities like CT in the field of "Medical Detection Systems (MDS)" because of their better contrast in muscle tissue in humans[13]. Currently, the majority of MRI anomaly identification is manual, and it takes more time for doctors for locating and segmenting tumours to surgical, therapeutic purposes[14] [15] [16]. This manual method might endanger life and is also prone to mistakes. To solve these issues, research has started to focus on various deep learning and machine learning techniques for computer-based cancer segmentation and diagnosis. 2. Related work In [19], they proposed a technique based on genetic algorithms (GA) and CNNs to noninvasively aggregate different assessments of glioma using MRI. They explained how a specific strategy was used to construct the CNN's architecture using GA as well as how stowing was used to reduce change in expectation error. Glioma is a form of brain tumour that is extremely lethal and has a lifespan of just two years. In [20], authors published automatically estimate the longevity of patients with glioma. The doctor gave the patient's projected number of days to live up until the patient was given the Glioma diagnosis. To avoid any human mistake, the technology automated the process for the impartial prediction. MRI imaging data with ground-truth segmented label identifies region-of-interest (ROI). In [21], author developed acquiring CNN highlights, such as images of the brain and liver, includes “discrete wavelet transform (DWT)” intensity, descent while signal processing, and included LSTM intensity for signal analyzing. To arrange those MRIs of brains with tumours and group CT images of livers among tumours, a CNN+LSTM+DWT method had been developed. Element vector pertaining to those pictures are obtained by configuring AlexNet architecture for the first half and second half as CNN+LSTM+DWT technique. For increasingly precise new disease detection, the author of [22] suggested a data processing method for maintaining data of the entire cortical covering inside deep networks. To categorise Alzheimer's disease, a 3D MRI of brain is used. Cortical covering is then straightened as 2D plane Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 333 https://internationalpubls.com and deep networks are placed over this 2D cortical covering. Leveled cortical pictures were used on several deep networks, including ResNet, together with ADNI collection of brain MRI filters. In [23], the author used reliable and conventional procedures to identify brain tumour, determine characteristic, categorize glioma by MRI. As a consequence, they created the model which uses synthetic neural networks and image processing to assist in the diagnosis of brain tumours. Primary picture has been made more complexity by using Histogram Balance (HE) method. In [24], they presented the MV-KBC deep technique, which uses limited chest CT data to distinguish dangerous from favourable knobs by multi-view knowledge-based collaboration. Their approach disassembles a 3D knob into nine predetermined views to study the features of the 3D lung knob. Authors created “knowledge-based collaborative (KBC)” subsystem to every scene, three different image types are used to change 3 different ResNet-50 systems which depict overall character, voxel, and form heterogeneity of knobs independently. Authors used 9 KBC submodels to describe lung knobs using the flexible weighting strategy learned by error back-propagation, allowing MV-KBC model to produce in the comprehensive manner. Role of punitive misfortune was applied to GA with minimal influence on overall MV-KBC execution in order to significantly reduce bogus erroneous valuation. Regarding brain tumour localization and containment, unique dual platforms multi- modelled program conclusion structure was provided by the author of [25]. The initial stage of framework construction includes pre-processing, feature extraction by CNN, and feature analyses by ECOC-SVM method. Goal is to identify brain tumours by categorising MRIs as common and unusual MRI. Purpose of succeeding system stage was to use the fully prepared 5 layer location- based R-CNN to limit tumour inside abnormal MRIs. They obtained an 87% on the dice[35][38]. Convolution neural networks are used in [26] to split brain tumours into smaller segments by breaking them out into short 3x3 kernels. Three distinct CNN strategies for glioma segmentation to MICCAI BraTS challenge against dataset were put up by the author in [27]. By using pre-ordered patterns from BraTS dataset for segmenting images from Rembrandt dataset, they looked at differences in the way information was received between several other datasets and BraTS dataset. The outcomes show that segmentation's only dice score is 86%[36][37]. In [28], CNN methods evaluated the effectiveness of using more under CNN topologies for brain tumour segmentation. By inserting a few 3 x 3 estimated channel into convolution phases, this process is demonstrated. As a result, more advanced convolution stages may be added to system without reducing robust open field of conventional upper channels. Additionally, deep designs employ more non-linearity and have lower channels load due to usage of smaller channels, which reduces the risk of overfitting. “Leaky rectified linear unit (LReLU)”, the ReLU altered model, was used as non-linearity initiating task. Rascal dice records of 88 percent, 83 percent, and 77 percent were obtained to whole tumour, center tumour, dynamic tumour, respectively, using suggested CNN which creates Eleven layers of profundity “6 six convolutional covers followed by 3 completely associated stages with two maximum pooling stages isolating them into squares of three”. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 334 https://internationalpubls.com Another novel strategy was used to a two-pathway CNN design in [29]. In order to figure out a fallen CNN that analyses neighbourhood nuances of brain MRI in addition to larger settings of brain tissues, smaller measured patches and larger estimated fixes must be extracted concurrently. In order to position the mark of the focus pixel, patches measuring 65 x 65 were extricated to globe pathway and patches measuring 33 x 33 pixel are extricated to local pathway from each individual MRI approach. Multi-method 2D design A CNN initially prepared globe input patches with 65 x 65 x 4 size to produce 33 × 33 x 5 sized patches. Later, these patches are connected to 33x33x4 neighbour patches which where sent as input to the 2 track CNN with convolution stages that had 7 x 7 estimation channels in one manner and 13 x 13 measure channels in preceding one. In [30], they built four separate CNNs' unique information-seeking bits from each single MRI technique picture by selecting several plane fixes inside every pixel and performing these steps. A RF classifier was prepared by deep stages of CNN, which are successfully linked, used as highlighted map. Only a precision stage of 67% was provided, with no information on the use of pre/post- handling levels. A limited architecture prediction with CNN is suggested in [31], in the overall approach. First fixes of name are taken from sands precision pattern and afterwards packaged by k-implies method into N gathering to form the marking fix lexicon of size N. This method replaces the use of CNNs to define focused voxels regarding data picture pieces inside brain tissue classes. The multi-levelled model elements inside unity of those assemblages are then characterized using a 2D CNN. Whelp cube rates of 83 percent, 75 percent, and 77 percent for the overall tumour, center tumour, dynamic tumour regions were independently taken into consideration in this segmentation execution of the technique. Another comparable strategy was published in [32] and mentioned in [33]. Two-stage preparation is also used in conjunction with this innovative compositional process to prevent class disparities in characters. CNN is retrained by expanding delegate diffusion of first metaphor during the subsequent stage after falling behind with its preparation of groups in the initial stage. The corresponding segment's technique was also implemented as a posthandling action and Maxout non-linearity was used. High Minxe cube rates 88 percent are applied to overall tumour region, 79% are applied to the centre tumour district, and 73% are applied to the dynamic tumour location. 3 Methodology Here, the methodology for the suggested system of brain tumour detection and classification is described. A more suitable deep learning model that is "Faster R-CNN (Faster Region-based Convolutional Neural Network) and combined with deep feature extractor: Residual Network-50 (ResNet-50)" to identify and categorize various brain tumours on the Kaggle brain tumour dataset [34]. 3.1 System Overview This study uses deep learning as major component of proposed system to identify 3 different forms of brain tumours, like pituitary, meningioma, glioma, which have an influence on the human brain. Figure 1 shows a broad view of the system. Below, each stage of the proposed system is thoroughly explained. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 335 https://internationalpubls.com 3.2 Data Collection A brain tumour dataset includes MRI pictures with top and side views of the brain for glioma, meningioma, pituitary, and no tumour. The symptoms of pituitary, meningioma, glioma, and samples of MRI are summarised in the table (Table 2). 3.3 Data Annotation All tumour locations were manually labelled in all photos using the LabelImg tool, which is an excellent tool for labelling images, by the bounding-boxes and class type to which the tumour belongs. The coordinates of various sized bounding boxes with their corresponding tumour class are the output of the data annotation step, and these coordinates will be assessed by the IoU (Intersection-over-Union) using the suggested system and projected outcome during testing. Figure 1. High-level picture of proposed system for brain tumour detection and classification Table 2. Dataset for brain tumour detection and classification Tumour Common Symptoms Collected sample Glioma tumour Seizures, behavioural changes, nausea and vomiting, hemiparesis, headaches, dizziness, difficulty walking, cognitive problems, vision loss, aphasia Meningioma tumour Language difficulty, changes in vision, headaches, hearing loss, memory loss, loss of smell, seizures, legs or arms weakness Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 336 https://internationalpubls.com Pituitary tumour Sudden weight gain or loss, Increased amount of urine, sexual dysfunction, less frequent or no menstrual periods, feeling cold, weakness, nausea and vomiting 3.4 Faster-RCNN+ResNet50 Brain Tumour Detection and Classification The objective is to identify three different groups of brain tumours and their locations in brain MRI scans. The system's bounding boxes that include the tumour must be accurately established in order to get reliable findings. Using Region Proposal Network (RPN), faster R-CNN produces crucial Region of Interest (ROI) for the identification of brain tumours. In a Faster RCNN technique for brain tumour detection in MRI, the following stages are carried out: 1. Provide an MRI to the ConvNet, which will then provide feature maps to RPN for the MRI. 2. RPN generates k fixed size anchor boxes in a brain tumour MRI by sliding windows on collected feature maps at each window, and predicts the likelihood that an anchor is a tumour as well as the bounding-box regressor that will most closely fit the anchor for a tumour. 3. After getting various types and sizes of bounding boxes and cropping each proposal such that it includes a brain tumour, ROI gathers fixed-size feature maps to all of the anchors. 4. The gathered fixed size feature mappings are sent to a fully connected layer that has a softmax and a linear regression layer. With the use of the ResNet 50 classifier, the brain tumour is ultimately classified as (glioma tumour, meningioma tumour, pituitary tumour, and no tumour), and bounding- boxes to detect the location of the brain tumour in an MRI are predicted. 3.5 RPN training and loss functions An anchor is considered a "positive (tumour)" sample if it satisfies one of the two requirements listed below: With a ground-truth box, which have the highest “Intersection over Union (IoU)”; when the IoU is greater than 0.5 with any ground-truth-box. Anchor is labelled as "negative (no tumour)" if IoU on other ground-truth-boxes is substantially lesser than 0.5. During RPN training, all anchors are disregarded (either negative or positive). A single MRI serves as the basis for each RPN mini-batch. A batch should be created with 128 negatives and 128 positive samples to prevent bias learning. The loss of RPN training is calculated by: Loss({ pᵢ }, { tᵢ })= 1/Ncls ∑i Losscls (pᵢ, pᵢ*)+ λ*(1/Nreg )∑i pᵢ*Lossreg(tᵢ, tᵢ*) (1) Here, Lossreg(tᵢ, tᵢ*) – Regression loss Losscls (pᵢ, pᵢ*) - Classification loss i - Anchor index in the mini-batch pᵢ - output score for anchor i from the classification branch pᵢ* - Ground truth label (0 or 1) Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 337 https://internationalpubls.com 4 Experiments and results Using labelImg, manually annotated all MRI in the brain tumour dataset, the total number of labels for class considered for the proposed system for training and testing is shown in table 3. Table 3. Total number of labels considered for training and testing for each class Class No. of labels for training No. of labels for testing Glioma tumour 826 100 Meningioma tumour 822 115 Pituitary tumour 827 74 No tumour 395 105 Total 2870 394 The suggested approach employs ResNet-50 and Faster R-CNN to identify three types of common brain tumours. First, IoU and AP (Average Precision) analyze the proposed system's performance. IoU(B, S) = │(B∩S)/(BUS)│ (3) Here, B – Annotated Coordinates of Ground-truth-box S – Predicted result by system When calculated IoU exceeds the threshold value of 0.5, projected result was classified as either the “true positive (TP) or the false positive (FP)”. When there are few FPs from MRI’s including tumour locations, the suggested approach is deemed effective. Mean Average Precision (mAP) is AP determined for glioma tumour, meningioma tumour, pituitary tumour, and no tumour. AP is generated by calculating precision average across the range of [0, 0.1,..., 1] recall levels. AP =1/11∑r Pinter (r) (4) Pinterp(r) = maxp(ř) (5) ř: ř≥r Here, p(ř) is accuracy measure at recall for each brain tumour. Suggested detection mechanism and mAP calculated to IoU = 0.5 obtained more than 94%. Table 4 displays the faster RCNN with ResNet50 findings for pituitary, meningioma and giloma tumours. Performance of suggested system, mAP, exceeded 94%. The performance can be improved by training proposed system with more samples. Figure 2 displays resultant loss curve for the fifty four thousand epochs and demonstrates that the proposed is capable of learning brain tumour data by achieving a smaller error-rate less than 0.1 at fifty two thousand epoch. Tensorboard is a tensorflow visualization toolkit that provides visualization and tracking of total loss. Table 5 displays the qualitative outcomes obtained by the suggested approach to identify three common brain tumours. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 338 https://internationalpubls.com Table 5. Qualitative results predicted by the proposed system Result predicted the system Whether prediction is correct or not Correct prediction Correct prediction Correct prediction Correct prediction Figure 2: Total loss curve of the proposed system Table 4. Proposed system results achieved for brain tumour detection Class ResNet50 Glioma tumour 96.22% Meningioma tumour 93.50% Pituitary tumour 92.09% x axis - Number of epoch’s y axis - Loss Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 339 https://internationalpubls.com No tumour 94.48% Total mean AP 94.07% 5. Conclusion Over 28,000 cases of brain tumours have been found, and more than 24,000 people in India lose their lives to brain tumours each year, according to the IARC. The increased use of electronic gadgets like tablets, smartphones, and other devices has increased the prevalence of this illness among kids as well. However, if a tumour is found and diagnosed early in the stage of tumour growth, there is a good chance that the patient will be treated. As a result, early tumour diagnosis is essential for successful treatment. The better deep learning model, "Faster R-CNN (Faster Region-based Convolutional Neural Network) and combined with deep feature extractor: Residual Network-50 (ResNet-50)" is suggested for the detection and classification of various brain tumours, including pituitary, meningioma, and glioma, on the brain tumour MRI dataset. For glioma, meningioma, pituitary, and no tumour, an MRI image collection of the brain's top and side views is available. The proposed system, achieved mAP 94%. References [1] Bhagat, Ms Jayali Vilas, and N. B. Dhaigude. "A Survey on Brain tumor Detection Tec hniques." International Research Journal of Engineering and Technology 4.3 (2017): 1795-1796. [2] Sharif, Muhammad Irfan, et al. "A comprehensive review on multi-organs tumor detection based on machine learning." Pattern Recognition Letters 131 (2020): 30-37. [3] Panda, Bichitra, and Chandra Sekhar Panda. "A review on brain tumor classification methodologies." International Journal of Scientific Research in Science and Technology (2019): 346-359. [4] Kaur, Ranjeet, and Amit Doegar. "Localization and classification of brain tumor using machine learning & deep learning techniques." International Journal of Innovative Technology and Exploring Engineering (IJITEE) 8.9S (2019). [5] Nazir, Muhammad, et al. "Brain tumor detection from MRI images using multi-level wavelets." 2019 international conference on Computer and Information Sciences (ICCIS). IEEE, 2019. [6] Kaur, Harminder, and Amninder Kaur Gill. "Review of Brain Tumor Detection Using Various Techniques." Int. J. Adv. Res. Comput. Sci. Softw. Eng. 7.5 (2017): 867-870. [7] Sharif, Muhammad, et al. "An integrated design of particle swarm optimization (PSO) with fusion of features for detection of brain tumor." Pattern Recognition Letters 129 (2020): 150-157. [8] https://health.economictimes.indiatimes.com/news/industry/world-brain-tumour-day-2022-together-we-are- stronger/92078118 [9] Sarhan, Ahmad M. "Brain tumor classification in magnetic resonance images using deep learning and wavelet transform." Journal of Biomedical Science and Engineering 13.06 (2020): 102. [10] Karuna, M., and Ankita Joshi. "Automatic detection and severity analysis of brain tumors using gui in matlab." International Journal of Research in Engineering and Technology 2.10 (2013): 586-594. [11] Aboody, Karen S., et al. "Neural stem cells display extensive tropism for pathology in adult brain: evidence from intracranial gliomas." Proceedings of the National Academy of Sciences 97.23 (2000): 12846-12851. [12] Joshi, Megha A., and D. H. Shah. "Survey of brain tumor detection techniques through MRI images." AIJRFANS, ISSN (2015): 2328-3785. [13] Dixit, Vinita, and Jyotika Pruthi. "Review of Image Processing Techniques for Automatic Detection of tumor in human liver." International journal of computer science and mobile computing 3.3 (2014): 371-378. [14] Abd-Ellah, Mahmoud Khaled, et al. "A review on brain tumor diagnosis from MRI images: Practical implications, key achievements, and lessons learned." Magnetic resonance imaging 61 (2019): 300-318. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 340 https://internationalpubls.com [15] Bhatele, Kirti Raj, and Sarita Singh Bhadauria. "Brain structural disorders detection and classification approaches: a review." Artificial Intelligence Review 53.5 (2020): 3349-3401. [16] Kumari, Nitu, and Sanjay Saxena. "Review of brain tumor segmentation and classification." 2018 International Conference on Current Trends towards Converging Technologies (ICCTCT). IEEE, 2018. [17] https://www.cancer.net/cancer-types/brain-tumor/risk-factors [18] https://www.cancer.net/cancer-types/brain-tumor/symptoms-and-signs [19] Anaraki, A. K., Ayati, M., & Kazemi, F. (2018). Magnetic resonance imaging-based brain tumor grades classification and grading via convolutional neural networks and genetic algorithms. Biocybernetics and Biomedical Engineering. j.bbe.2018.10.004 [20] Sharmila A. A, Arun D, Venkatesh J, Sudarshan S, Pranav A, "Predicting Sur-vival of Brain Tumor Patients using Deep Learning" International Journal of Innovative Technology and Exploring Engineering (IJITEE) ISSN: 2278- 3075, Volume-8 Issue-6, April, 2019. [21] Hüseyin Kutlu and Engin Avg. A Novel Method for Classifying Liver and Brain Tumors Using Convolutional Neural Networks, DiscreteWavelet Trans-form, and Long Short-Term Memory Networks. Sensors, 19, 1992, 2019. [22] Menikdiwela, M., Nguyen, C., & Shaw, M. (2018). Deep Learning on Brain Cortical Thickness Data for Disease Classification. 2018 Digital Image Com-puting: Techniques and Applications (DICTA). dicta.2018.8615775. [23] Niyaz, U., Sambyal, A. S., & Devanand. (2018). Advances in Deep Learning Techniques for Medical Image Analysis. 2018 Fifth International Conference on Parallel, Distributed and Grid Computing (PDGC). pdgc.2018.8745790 [24] Xie, Y., Xia, Y., Zhang, J., Song, Y., Feng, D., Fulham, M., & Cai, W. (2018). Knowledge-based Collaborative Deep Learning for Benign-Malignant Lung Nodule Classification on Chest CT. IEEE Transactions on Medical Imaging, tmi.2018.2876510 [25] Mahmoud Khaled Abd-Ellah, Ali Ismail Awad, Ashraf A. M. Khalaf and Hesham F. A. Hamed. Two-phase multi- model automatic brain tumor diagnosis system from magnetic resonance images using convolutional neural net- works. EURASIP Journal on Image and Video Processing (2018) 2018:97. [26] Manda SSSNMSRL Pavan, and P. Jagadeesh. Brain Tumor Segmentation Using Convolutional Neural Network in MRI Images. International Journal of Pure and Applied Mathematics, Volume 119, No. 17, 2018, 1585-1592. [27] Balaji, K., P. Sai Kiran, and M. Sunil Kumar. "Power aware virtual machine placement in IaaS cloud using discrete firefly algorithm." Applied Nanoscience 13.3 (2023): 2003-2011. [28] Balaji, K., P. Sai Kiran, and M. Sunil Kumar. "Resource aware virtual machine placement in IaaS cloud using bio- inspired firefly algorithm." Journal of Green Engineering 10 (2020): 9315-9327. [29] Sangamithra, B., B. E. Manjunath Swamy, and M. Sunil Kumar. "Personalized Ranking Mechanism Using Yandex Dataset on Machine Learning Approaches." In Proceedings of the International Conference on Cognitive and Intelligent Computing: ICCIC 2021, Volume 1, pp. 629-639. Singapore: Springer Nature Singapore, 2022. [30] Ganesh, D., Rao, K. J., Kumar, M. S., Vinitha, M., Anitha, M., Likith, S. S., & Taralitha, R. (2023, March). Implementation of Novel Machine Learning Methods for Analysis and Detection of Fake Reviews in Social Media. In 2023 International Conference on Sustainable Computing and Data Communication Systems (ICSCDS) (pp. 243-250). IEEE. [31] Burada, Sreedhar, BE Manjunath Swamy, and M. Sunil Kumar. "Early Detection of Melanoma Skin Cancer: A Hybrid Approach Using Fuzzy C-Means Clustering and Differential Evolution-based Convolutional Neural Network." Measurement: Sensors (2024): 101168. [32] Davy A. et al. Brain tumor segmentation with deep neural networks. MICCAI Multimodal Brain Tumor Segmentation Challenge (BraTS) 2014:1–5. [33Burada, S., Eraiah, M. B., & Kumar, M. S. (2024). Optimal hybrid classifier with fine-tuned hyper parameter and improved fuzzy C means segmentation: skin cancer detection. International Journal of Ad Hoc and Ubiquitous Computing, 45(1), 52-64. [34] https://www.kaggle.com/datasets/sartajbhuvaji/brain-tumor-classification-mri https://www.kaggle.com/datasets/sartajbhuvaji/brain-tumor-classification-mri Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 31 No. 6s (2024) 341 https://internationalpubls.com [35] Natarajan, V. Anantha, Ms Macha Babitha, and M. Sunil Kumar. "Detection of disease in tomato plant using Deep Learning Techniques." International Journal of Modern Agriculture 9, no. 4 (2020): 525-540. [36] Natarajan, V. A., Kumar, M. S., Patan, R., Kallam, S., & Mohamed, M. Y. N. (2020, September). Segmentation of nuclei in histopathology images using fully convolutional deep neural architecture. In 2020 International Conference on computing and information technology (ICCIT-1441) (pp. 1-7). IEEE. [37] Sreedhar, B., BE, M. S., & Kumar, M. S. (2020, October). A comparative study of melanoma skin cancer detection in traditional and current image processing techniques. In 2020 Fourth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I-SMAC) (pp. 654-658). IEEE. [38] Natarajan, M. K. V. A., & Macha Babitha, D. G. Machine Learning Based Identification of Covid-19 From Lung Segmented CT Images Using Radiomics Features. Biosc. Biotech. Res. Comm. Special Issue, 14(07), 350-355.