SlideShare a Scribd company logo

Brain Tumor Detection using CNN

M
MohammadRakib8

The document describes a study that used convolutional neural networks (CNNs) to detect brain tumors in MRI images. Three CNN models were developed and their performance was evaluated using metrics like accuracy, precision, recall, F1-score, and confusion matrices. Model 3 achieved the highest test accuracy of 94% for tumor detection. In total, over 2000 MRI images were used in the study after data augmentation. The CNN models incorporated convolution, pooling, and fully connected layers to analyze image features and classify tumors. This research demonstrates that CNNs can accurately detect brain tumors in medical images.

Education
1 of 8
Download to read offline
1
2
Most read
3
Most read
4
5
6
7
8
Most read
Brain Tumor Detection using CNN Drubojit Saha 170104027@aust.edu Mohammad Rakib-Uz-Zaman 170104041@aust,edu Tasnim Nusrat Hasan 170104046@aust.edu Abstract—Diagnosis of brain tumors is an essential task in the medical field for finding out if the tumor can probably become cancerous. Deep learning is a convenient and decisive approach for image classification. It has been broadly applied in diverse fields like medical imaging, as its application does not require the reliability of a skilled in the related field, but rather requires the number of data and distinct data to produce good classification conclusions. Convolutional Neural Network (CNN) is used for image classification as well as recognized because of its immense accuracy. In this paper, a comparison between two models of CNN of our selected paper is shown to find the best model to classify tumors in Brain MRI Image and at last, another approach of a CNN model is trained and gained a prediction accuracy of up to 94%. I. INTRODUCTION Medical imaging invokes a number of techniques that can be used as non-interfering methods of looking inside the body. Medical image compasses various image modalities and converts to image of the human body for analyzing and investigating purposes and thus it plays a great and decisive role in taking actions for the enhancement of people’s health. Image segmentation is an essential stride in image process- ing which actuates the accomplishment of a higher level of image processing. The fundamental goal of image segmenta- tion in medical image processing is mainly detecting tumors, competent machine vision and gaining satisfactory result for further diagnosis. Brain, as well as other nervous system cancer, is the 10th leading reason of death and the five-year endurance rate for people with a cancerous brain is 34 percent for men on the other hand 36 percent for women. The World Health Organization (WHO) states that around 400,000 people around the world are affected by the brain tumor and 120,000 people have died in the recent years. Early detection of brain tumors has played an imperative role in developing the treatment possibilities, and a higher gain of survival possibility can be achieved. Although manual segmentation of tumors is a tedious, challenging, and difficult task as it requires a large number of MRI images that are generated in medical routine. Magnetic Resonance Imaging is mainly used for brain tumor detection. Brain tumor seg- mentation from MRI is one of the most compelling tasks in medical image processing as it involves a considerable amount of data. Furthermore, the tumors can be ill-defined with soft tissue boundaries. As a result, it is a very comprehensive task Fig. 1. Dummy Input and Output of Proposed System to gain the accurate segmentation of tumors from the human brain. II. RELATED WORKS In [1] an achievement of substantial results in image segmentation and classification is shown through the convolutional neural network (CNN). A new CNN architecture for brain tumor classification network is simpler than already-existing pre-trained networks, and it was tested on contrast-enhanced magnetic resonance images. In [2] they have established the whole segmentation process based on Mathematical Morphological Operations and applied spatial FCM algorithm which improves the computation time. In [3] they have established that Convolutional Neural Networks are good enough to diagnose brain tumors on MRI images. This study resulted in accuracy of 93% and a loss value of 0.23264. The amount of convolution layers that affects the quality of classification, more convolution layers rise the accuracy results, but more convolution layers will require more time for training. III. OBJECTIVE Convolutional Neural Networks are widely used in the medical field for image processing. Over the years lots after trying researchers built a model which can detect the tumor more accurately. In this paper, the same idea has come up which can accurately identify the tumor from Brain MRI images. A fully connected neural network layer can detect the tumor, but due to parameter sharing and sparsity of connection, CNN is used as our model. The dummy figure of CNN is shown in Figure 1.
Fig. 2. Proposed Method A. Convolution Layer Convolution Layer is the main layer in the CNN method which targets to extract features from the input. Convolution executes linear transformations of input data outwardly chang- ing spatial information in the data. Convolution kernels are regulated from the weight of the layer in case the convolution kernels can process the input data training on CNN. B. Subsampling Layer Subsampling focuses on reducing the size of image data and enhancing the invariance of feature positions. CNN applies Max Pooling as a subsampling method. The way Max Pooling works is to break the output of the convolution layer into several smaller grids and after that take the maximum value from each grid to construct a smaller image matrix. Small image size will make it easier to process the next convolution layer C. Fully Connected Layer The Fully Connected Layer evolves the dimensions of the data so that it can be classified linearly. In the convolution layer, each neuron must be converted into one-dimensional data before including into another layer which is connected as a whole. This process is induced by data losing its spatial information and at last Fully Connected Layer network is applied. IV. METHODOLOGY A. Proposed Method This paper implements CNN for the detection of brain tumors. This study accepts input images labeled as yes or no from the raw dataset and then applies these patterns to categorize between tissues that do not contain tumors and those that contain tumors. This paper has implemented 2 model for CNN. An extra model of CNN is also implemented afterwards. Therefore, the proposed system is illustrated in Figure 2. B. Data Augmentation The amount of data in the dataset is not sufficient to be used as training data for CNN. As a result, the augmentation method is used to overcome the imbalance of issues. Augmentation is an algorithm that can utilize statistical data information and form an integrated model. Fig. 3. First CNN Model Fig. 4. Second CNN Model C. Image Pre-processing Pre-processing is executed to create smooth training as there are different variants of intensity, contrast, and size in images. The proportion of the images are resized to shape (240, 240, 3) = (imagewidth, imageheight, numberofchannels) because the images in the dataset have different sizes. After normalization, scale pixel values to the range 0-1 to facilitate the learning process. D. Model CNN In this research, the CNN model contains several layers, namely the convolution layer, the pooling layer, the dense layer, the flatten layer, and the dropout layer. Along with the layers used in the CNN process, there is also an activation function in this study using Relu activation. In this paper, 3 CNN models are made as comparison material. The CNN model design can be seen in Figure 3, Figure 4 and Figure 5. An image in the form of a total association with the first convolution, image size of 240x240 pixels. Kernels have a size of 3x3 and filters are used as many as 32. After that, the model will perform the activation and pooling data functions. The Pooling layer process handles to diminish the dimensions of the feature map. The derivation of the convolution process is mainly a feature map that is employed for the consecutive convolution process repeatedly. The next step is a flatten feature map in vector form to carry out a fully connected layer to produce a classification of images. V. EXPERIMENTS A. Dataset The dataset used in this study is Brain MRI Images for Brain Tumor Detection obtained from kaggle.com. The dataset consists of 253 images assembled into 2 groups, 155 brain
Fig. 5. Third CNN Model Fig. 6. First Image without having any Tumor and Second Image having a Brain Tumor images that have tumors, and the rest 98 brain images that do not have tumors. After data augmentation, the dataset subsists of 1085 samples containing tumors and 980 samples not containing tumors, bringing a total of 2065 images. The images having tumor and no tumor is shown in Figure 6. The dataset is splitted into training, testing and validation set with 70:15:15 ratio. B. Evaluation Metrices The evaluation matrices for the three model are repre- sented by Classification Report(Accuracy, Precision, Recall, F1-score), Confusion Matrix as well as Loss Curve. 1) CNN Model 1: For Model 1 the evaluation matrices are shown in Figure 7. Figure 8, Figure 9. Figure 10, Figure 11. Figure 12, Figure 13. Figure 14, Figure 15. Figure 16, Figure 17. Figure 18, Figure 19. Figure 20, Figure 21. Figure 22, Figure 23. Figure 24, Figure 25. Figure 26. Fig. 7. Confusion Matrix and Loss Curve of Model 1 (Run 1) Fig. 8. Classification Report of Model 1 (Run 1) Fig. 9. Confusion Matrix and Loss Curve of Model 1 (Run 2) Fig. 10. Classification Report of Model 1 (Run 2) Fig. 11. Confusion Matrix and Loss Curve of Model 1 (Run 3) Fig. 12. Classification Report of Model 1 (Run 3) Fig. 13. Confusion Matrix and Loss Curve of Model 1 (Run 4) Fig. 14. Classification Report of Model 1 (Run 4) Fig. 15. Confusion Matrix and Loss Curve of Model 1 (Run 5)
Fig. 16. Classification Report of Model 1 (Run 5) Fig. 17. Confusion Matrix and Loss Curve of Model 1 (Run 6) Fig. 18. Classification Report of Model 1 (Run 6) Fig. 19. Confusion Matrix and Loss Curve of Model 1 (Run 7) Fig. 20. Classification Report of Model 1 (Run 7) Fig. 21. Confusion Matrix and Loss Curve of Model 1 (Run 8) 2) CNN Model 2: For Model 2, the evaluation matrices are shown in Figure 27. Figure 28, Figure 29. Figure 30, Figure 31. Figure 32, Figure 33. Figure 34, Figure 35. Figure 36, Figure 37. Figure 38, Figure 39. Figure 40, Figure 41. Figure 42, Figure 43. Figure 44, Figure 45. Figure 46. 3) CNN Model 3: For Model 2, the evaluation matrices are shown in Figure 47. Figure 48, Figure 49. Figure 50, Figure Fig. 22. Classification Report of Model 1 (Run 8) Fig. 23. Confusion Matrix and Loss Curve of Model 1 (Run 9) Fig. 24. Classification Report of Model 1 (Run 9) Fig. 25. Confusion Matrix and Loss Curve of Model 1 (Run 10) Fig. 26. Classification Report of Model 1 (Run 10) Fig. 27. Confusion Matrix and Loss Curve of Model 2 (Run 1) Fig. 28. Classification Report of Model 2 (Run 1)
Fig. 29. Confusion Matrix and Loss Curve of Model 2 (Run 2) Fig. 30. Classification Report of Model 2 (Run 2) Fig. 31. Confusion Matrix and Loss Curve of Model 2 (Run 3) Fig. 32. Classification Report of Model 2 (Run 3) Fig. 33. Confusion Matrix and Loss Curve of Model 2 (Run 4) Fig. 34. Classification Report of Model 2 (Run 4) Fig. 35. Confusion Matrix and Loss Curve of Model 2 (Run 5) Fig. 36. Classification Report of Model 2 (Run 5) Fig. 37. Confusion Matrix and Loss Curve of Model 2 (Run 6) Fig. 38. Classification Report of Model 2 (Run 6) Fig. 39. Confusion Matrix and Loss Curve of Model 2 (Run 7) Fig. 40. Classification Report of Model 2 (Run 7) Fig. 41. Confusion Matrix and Loss Curve of Model 2 (Run 8) Fig. 42. Classification Report of Model 2 (Run 8)
Fig. 43. Confusion Matrix and Loss Curve of Model 2 (Run 9) Fig. 44. Classification Report of Model 2 (Run 9) Fig. 45. Confusion Matrix and Loss Curve of Model 2 (Run 10) Fig. 46. Classification Report of Model 2 (Run 10) 51. Figure 52, Figure 53. Figure 54, Figure 55. Figure 56, Figure 57. Figure 58, Figure 59. Figure 60, Figure 61. Figure 62, Figure 63. Figure 64, Figure 65. Figure 66. Fig. 47. Confusion Matrix and Loss Curve of Model 3 (Run 1) Fig. 48. Classification Report of Model 3 (Run 1) Fig. 49. Confusion Matrix and Loss Curve of Model 3 (Run 2) Fig. 50. Classification Report of Model 3 (Run 2) Fig. 51. Confusion Matrix and Loss Curve of Model 3 (Run 3) Fig. 52. Classification Report of Model 3 (Run 3) Fig. 53. Confusion Matrix and Loss Curve of Model 3 (Run 4) Fig. 54. Classification Report of Model 3 (Run 4) Fig. 55. Confusion Matrix and Loss Curve of Model 3 (Run 5)
Fig. 56. Classification Report of Model 3 (Run 5) Fig. 57. Confusion Matrix and Loss Curve of Model 3 (Run 6) Fig. 58. Classification Report of Model 3 (Run 6) Fig. 59. Confusion Matrix and Loss Curve of Model 3 (Run 7) Fig. 60. Classification Report of Model 3 (Run 7) Fig. 61. Confusion Matrix and Loss Curve of Model 3 (Run 8) C. Results Experiments in this article are carried out on 2065 images consisting of 1085 samples containing tumors and 980 samples containing no tumors. The data has been run for 10 times, each using the CNN model that has been made before, each experiment using 25 epochs and 32 batches. Fig. 62. Classification Report of Model 3 (Run 8) Fig. 63. Confusion Matrix and Loss Curve of Model 3 (Run 9) Fig. 64. Classification Report of Model 3 (Run 9) Fig. 65. Confusion Matrix and Loss Curve of Model 3 (Run 10) Fig. 66. Classification Report of Model 3 (Run 10) 1) CNN Model 1: This model gives highest training accu- racy value of 100% and has lowest loss value of 0.002 on the training data, but there is a significant difference with the test data results, the highest test data accuracy value is 83% and the value of F1-Score is 0.825 presented in Table I. 2) CNN Model 2: This model gives highest training accu- racy value of 99.73% and lowest loss value of 0.0.0148, then the test data obtained a highest accuracy value of 90% and F1-Score value of 0.9 presented in Table II. 3) CNN Model 3: This model gives highest accuracy value of 95.64% and lowest loss value of 0.1101, then the test data obtained a highest accuracy value of 94% and F1-Score value of 0.94 presented in Table III.
TABLE I RESULT OF FIRST CNN MODEL No Train Validation Test Loss Accuracy Loss Accuracy Accuracy F1-score 1 .0033 1.0 0.9676 .822 0.83 0.825 2 .0076 1.0 0.9521 .7879 0.78 0.78 3 .0042 1.0 0.8097 0.8068 0.8 0.8 4 0.0387 0.996 0.725 0.7689 0.79 0.79 5 0.0094 1.0 0.8585 0.803 0.82 0.815 6 0.0062 1.0 0.8314 0.8068 0.8 0.8 7 0.0042 1.0 0.8243 0.7727 0.8 0.8 8 0.004 1.0 0.8259 0.8106 0.79 0.79 9 0.004 1.0 0.7821 0.7992 0.79 0.795 10 0.002 1.0 0.9391 0.7879 0.8 0.8 Average .00836 0.9996 0.8516 0.7966 0.8 0.7995 TABLE II RESULT OF SECOND CNN MODEL No Train Validation Test Loss Accuracy Loss Accuracy Accuracy F1-score 1 0.0179 0.9953 0.6348 0.8447 0.89 0.885 2 0.202 0.9933 0.3601 0.9129 0.87 0.87 3 0.0267 0.992 0.4388 0.8712 0.88 0.885 4 0.0438 0.9879 0.3646 0.9053 0.89 0.89 5 0.0517 0.9785 0.3507 0.8788 0.89 0.885 6 0.0793 0.9705 0.3205 0.9091 0.89 0.89 7 0.0379 0.9873 0.3942 0.8939 0.88 0.88 8 0.022 0.9933 0.3473 0.9129 0.89 0.89 9 0.0148 0.9973 0.359 0.875 0.88 0.88 10 0.0249 0.9906 0.3892 0.8939 0.9 0.9 Average 0.03392 0.9886 0.396 0.8898 0.886 0.8855 TABLE III RESULT OF THIRD CNN MODEL No Train Validation Test Loss Accuracy Loss Accuracy Accuracy F1-score 1 0.1171 0.9564 0.2091 0.9242 0.94 0.93 2 0.1627 0.9430 0.2242 0.9015 0.94 0.935 3 0.1342 0.9517 0.2203 0.9318 0.94 0.94 4 0.1772 0.9316 0.2982 0.8864 0.9031 0.905 5 0.1485 0.9416 0.3182 0.8750 0.91 0.91 6 0.1326 0.9484 0.2013 0.9318 0.9354 0.93 7 0.1623 0.9437 0.221 0.9318 0.93 0.93 8 0.1453 0.9437 0.2316 0.9053 0.94 0.935 9 0.1319 0.9504 0.2377 0.8977 0.9161 0.915 10 0.1101 0.9544 0.2105 0.9242 0.9129 0.91 Average 0.1422 0.9465 0.2372 0.911 0.9268 0.924 VI. CONCLUSION AND FUTURE DIRECTIONS Convolutional Neural Networks are great enough to diag- nose brain tumors using MRI images. This study resulted in an accuracy of 94% . The count of convolution layers affects the quality of classification shown in Figure 67 as more convolution layers expand the accuracy of results. The process of image augmentation can develop the alternatives of existing datasets, thereby raising the classification results. As this paper is developed by only using CNN, in future, this paper will be developed by using other hybrid deep learning algorithm. Fig. 67. Comparison of Models REFERENCES [1] Milica M Badža and Marko Č Barjaktarović. Classification of brain tumors from mri images using a convolutional neural network. Applied Sciences, 10(6):1999, 2020. [2] B Devkota, Abeer Alsadoon, PWC Prasad, AK Singh, and A Elchouemi. Image segmentation for early stage brain tumor detection using mathemat- ical morphological reconstruction. Procedia Computer Science, 125:115– 123, 2018. [3] DC Febrianto, I Soesanti, and HA Nugroho. Convolutional neural network for brain tumor detection. In IOP Conference Series: Materials Science and Engineering, volume 771, page 012031. IOP Publishing, 2020.

More Related Content

PPTX
Neural Network Based Brain Tumor Detection using MR Images
Aisha Kalsoom
 
PDF
MULTI-CLASSIFICATION OF BRAIN TUMOR IMAGES USING DEEP NEURAL NETWORK
Benyamin Moadab
 
PPTX
BRAIN TUMOR MRI IMAGE SEGMENTATION AND DETECTION IN IMAGE PROCESSING
Dharshika Shreeganesh
 
PPTX
Brain Tumor Detection Using Deep Neural Network.pptx
AbdulRehman417114
 
PPTX
Application of-image-segmentation-in-brain-tumor-detection
Myat Myint Zu Thin
 
PPTX
brain tumor ppt.pptx
AdityaSingh728086
 
PPTX
Brain tumor detection ppt (1)today.pptx
PoorabKumar
 
PPTX
PPT on BRAIN TUMOR detection in MRI images based on IMAGE SEGMENTATION
khanam22
 
Neural Network Based Brain Tumor Detection using MR Images
Aisha Kalsoom
 
MULTI-CLASSIFICATION OF BRAIN TUMOR IMAGES USING DEEP NEURAL NETWORK
Benyamin Moadab
 
BRAIN TUMOR MRI IMAGE SEGMENTATION AND DETECTION IN IMAGE PROCESSING
Dharshika Shreeganesh
 
Brain Tumor Detection Using Deep Neural Network.pptx
AbdulRehman417114
 
Application of-image-segmentation-in-brain-tumor-detection
Myat Myint Zu Thin
 
brain tumor ppt.pptx
AdityaSingh728086
 
Brain tumor detection ppt (1)today.pptx
PoorabKumar
 
PPT on BRAIN TUMOR detection in MRI images based on IMAGE SEGMENTATION
khanam22
 

What's hot (20)

PPT
Brain tumor detection by scanning MRI images (using filtering techniques)
Vivek reddy
 
PPTX
Brain tumor detection using convolutional neural network
MD Abdullah Al Nasim
 
PPTX
Brain Tumor Detection Using Image Processing
Sinbad Konick
 
PDF
Medical image analysis
Aboul Ella Hassanien
 
PPTX
Brain Tumour Detection.pptx
RevolverRaja2
 
PPTX
Final ppt
NijuMathew3
 
PDF
Lung Cancer Detection using transfer learning.pptx.pdf
jagan477830
 
PPTX
Convolutional neural network
MojammilHusain
 
PDF
Skin Cancer Detection Using Deep Learning Techniques
IRJET Journal
 
PPTX
Brain tumor detection using image segmentation ppt
Roshini Vijayakumar
 
PPT
Chapter10 image segmentation
asodariyabhavesh
 
PPT
Image inpainting
Pulkit Goyal
 
PPT
Tumour detection
Keerthi Kancharla
 
PPTX
IMAGE SEGMENTATION.
Tawose Olamide Timothy
 
PPTX
Deep learning for object detection
Wenjing Chen
 
PPTX
Lecture 1 for Digital Image Processing (2nd Edition)
Moe Moe Myint
 
PPTX
Brain Tumor Segmentation using Enhanced U-Net Model with Empirical Analysis
MD Abdullah Al Nasim
 
POTX
Presentation of Lossy compression
Omar Ghazi
 
PDF
Intro to Deep Learning for Medical Image Analysis, with Dan Lee from Dentuit AI
Seth Grimes
 
PPTX
Image Segmentation Using Deep Learning : A survey
NUPUR YADAV
 
Brain tumor detection by scanning MRI images (using filtering techniques)
Vivek reddy
 
Brain tumor detection using convolutional neural network
MD Abdullah Al Nasim
 
Brain Tumor Detection Using Image Processing
Sinbad Konick
 
Medical image analysis
Aboul Ella Hassanien
 
Brain Tumour Detection.pptx
RevolverRaja2
 
Final ppt
NijuMathew3
 
Lung Cancer Detection using transfer learning.pptx.pdf
jagan477830
 
Convolutional neural network
MojammilHusain
 
Skin Cancer Detection Using Deep Learning Techniques
IRJET Journal
 
Brain tumor detection using image segmentation ppt
Roshini Vijayakumar
 
Chapter10 image segmentation
asodariyabhavesh
 
Image inpainting
Pulkit Goyal
 
Tumour detection
Keerthi Kancharla
 
IMAGE SEGMENTATION.
Tawose Olamide Timothy
 
Deep learning for object detection
Wenjing Chen
 
Lecture 1 for Digital Image Processing (2nd Edition)
Moe Moe Myint
 
Brain Tumor Segmentation using Enhanced U-Net Model with Empirical Analysis
MD Abdullah Al Nasim
 
Presentation of Lossy compression
Omar Ghazi
 
Intro to Deep Learning for Medical Image Analysis, with Dan Lee from Dentuit AI
Seth Grimes
 
Image Segmentation Using Deep Learning : A survey
NUPUR YADAV
 
Ad

Similar to Brain Tumor Detection using CNN (20)

PDF
Convolutional Neural Network Based Method for Accurate Brain Tumor Detection ...
IRJET Journal
 
PPTX
Brain Tumor Detection Using Deep Learning ppt new made.pptx
vikyt2211
 
PPTX
Diagnosis with Medical Imaging by Slidesgo [Autosaved].pptx
XEON14
 
PDF
ugc carelist journals ugc carelist journals
mounikadopenventio
 
PDF
ugc carelist journals ugc carelist journals
mounikadopenventio
 
PDF
ugc carelist journals ugc carelist journals
mounikadopenventio
 
PDF
Brain Tumor Detection using Neural Network
ijtsrd
 
PDF
IRJET- Image Classification using Deep Learning Neural Networks for Brain...
IRJET Journal
 
PDF
Automated diagnosis of brain tumor classification and segmentation of magneti...
IAESIJAI
 
PDF
IRJET- Brain MRI Image Processing & Prediction of Cancer Stage Including ...
IRJET Journal
 
PDF
BRAIN TUMOR DETECTION
IRJET Journal
 
PDF
DIRECTIONAL CLASSIFICATION OF BRAIN TUMOR IMAGES FROM MRI USING CNN-BASED DEE...
IRJET Journal
 
PPTX
SHORT STORY_CMPE255.pptx
HarikaSatti1
 
PDF
Screening Brain Tumors from MRI Imagesw with Deep Learning Approaches
CSEIJJournal
 
PDF
Survey on “Brain Tumor Detection Using Deep Learning
IRJET Journal
 
PDF
Brain tumor classification in magnetic resonance imaging images using convol...
IJECEIAES
 
PDF
Hybrid model for detection of brain tumor using convolution neural networks
CSITiaesprime
 
PDF
ENHANCE BRAIN TUMOR DETECTION BY USING CNN AND AI TECHNIQUE
shashankbhandarkar91
 
PDF
IRJET- Brain Tumor Detection using Deep Learning
IRJET Journal
 
PPTX
INTEGRATION OF DEEP LEARNING AND IOT FOR REAL TIME BRAIN TUMOR DETECTION.pptx
amruthramesh1
 
Convolutional Neural Network Based Method for Accurate Brain Tumor Detection ...
IRJET Journal
 
Brain Tumor Detection Using Deep Learning ppt new made.pptx
vikyt2211
 
Diagnosis with Medical Imaging by Slidesgo [Autosaved].pptx
XEON14
 
ugc carelist journals ugc carelist journals
mounikadopenventio
 
ugc carelist journals ugc carelist journals
mounikadopenventio
 
ugc carelist journals ugc carelist journals
mounikadopenventio
 
Brain Tumor Detection using Neural Network
ijtsrd
 
IRJET- Image Classification using Deep Learning Neural Networks for Brain...
IRJET Journal
 
Automated diagnosis of brain tumor classification and segmentation of magneti...
IAESIJAI
 
IRJET- Brain MRI Image Processing & Prediction of Cancer Stage Including ...
IRJET Journal
 
BRAIN TUMOR DETECTION
IRJET Journal
 
DIRECTIONAL CLASSIFICATION OF BRAIN TUMOR IMAGES FROM MRI USING CNN-BASED DEE...
IRJET Journal
 
SHORT STORY_CMPE255.pptx
HarikaSatti1
 
Screening Brain Tumors from MRI Imagesw with Deep Learning Approaches
CSEIJJournal
 
Survey on “Brain Tumor Detection Using Deep Learning
IRJET Journal
 
Brain tumor classification in magnetic resonance imaging images using convol...
IJECEIAES
 
Hybrid model for detection of brain tumor using convolution neural networks
CSITiaesprime
 
ENHANCE BRAIN TUMOR DETECTION BY USING CNN AND AI TECHNIQUE
shashankbhandarkar91
 
IRJET- Brain Tumor Detection using Deep Learning
IRJET Journal
 
INTEGRATION OF DEEP LEARNING AND IOT FOR REAL TIME BRAIN TUMOR DETECTION.pptx
amruthramesh1
 
Ad

Recently uploaded (20)

PPTX
Cell Structure & Organelles in detailed.
DHANASHREESIVAKUMAR
 
PPTX
Final Presentation General Medicine 03-08-2024.pptx
ZaheerAhmad228692
 
PPTX
PPH.pptx obstetrics and gynecology in nursing
SrideviDevaraj5
 
PPTX
GDM (1) (1).pptx small presentation for students
shrawanbpkihs
 
PDF
Classroom Observation Tools for Teachers
Nicaramirez
 
PDF
Physiotherapy_for_Respiratory_and_Cardiac_Problems WEBBER.pdf
DrMONISHAphysio
 
PPTX
Lesson notes of climatology university.
sgladness67
 
PPTX
PPT- ENG7_QUARTER1_LESSON1_WEEK1. IMAGERY -DESCRIPTIONS pptx.pptx
KMDee
 
PDF
FourierSeries-QuestionsWithAnswers(Part-A).pdf
Raji Murugan
 
PDF
102 student loan defaulters named and shamed – Is someone you know on the list?
Kweku Zurek
 
PDF
O5-L3 Freight Transport Ops (International) V1.pdf
labyankof
 
PPTX
1st Inaugural Professorial Lecture held on 19th February 2020 (Governance and...
kawisojames10
 
PDF
TR - Agricultural Crops Production NC III.pdf
avgilmegino3
 
PDF
Microbial disease of the cardiovascular and lymphatic systems
KeyaSharma
 
PDF
Computing-Curriculum for Schools in Ghana
abigailanane203
 
PPTX
master seminar digital applications in india
ananyaray35
 
PPTX
school management -TNTEU- B.Ed., Semester II Unit 1.pptx
NihumathunnisaH
 
PDF
STATICS OF THE RIGID BODIES Hibbelers.pdf
pascualasturiaskaye4
 
PDF
Supply Chain Operations Speaking Notes -ICLT Program
labyankof
 
PDF
Basic Mud Logging Guide for educational purpose
wdandworks
 
Cell Structure & Organelles in detailed.
DHANASHREESIVAKUMAR
 
Final Presentation General Medicine 03-08-2024.pptx
ZaheerAhmad228692
 
PPH.pptx obstetrics and gynecology in nursing
SrideviDevaraj5
 
GDM (1) (1).pptx small presentation for students
shrawanbpkihs
 
Classroom Observation Tools for Teachers
Nicaramirez
 
Physiotherapy_for_Respiratory_and_Cardiac_Problems WEBBER.pdf
DrMONISHAphysio
 
Lesson notes of climatology university.
sgladness67
 
PPT- ENG7_QUARTER1_LESSON1_WEEK1. IMAGERY -DESCRIPTIONS pptx.pptx
KMDee
 
FourierSeries-QuestionsWithAnswers(Part-A).pdf
Raji Murugan
 
102 student loan defaulters named and shamed – Is someone you know on the list?
Kweku Zurek
 
O5-L3 Freight Transport Ops (International) V1.pdf
labyankof
 
1st Inaugural Professorial Lecture held on 19th February 2020 (Governance and...
kawisojames10
 
TR - Agricultural Crops Production NC III.pdf
avgilmegino3
 
Microbial disease of the cardiovascular and lymphatic systems
KeyaSharma
 
Computing-Curriculum for Schools in Ghana
abigailanane203
 
master seminar digital applications in india
ananyaray35
 
school management -TNTEU- B.Ed., Semester II Unit 1.pptx
NihumathunnisaH
 
STATICS OF THE RIGID BODIES Hibbelers.pdf
pascualasturiaskaye4
 
Supply Chain Operations Speaking Notes -ICLT Program
labyankof
 
Basic Mud Logging Guide for educational purpose
wdandworks
 

Brain Tumor Detection using CNN

  • 1. Brain Tumor Detection using CNN Drubojit Saha 170104027@aust.edu Mohammad Rakib-Uz-Zaman 170104041@aust,edu Tasnim Nusrat Hasan 170104046@aust.edu Abstract—Diagnosis of brain tumors is an essential task in the medical field for finding out if the tumor can probably become cancerous. Deep learning is a convenient and decisive approach for image classification. It has been broadly applied in diverse fields like medical imaging, as its application does not require the reliability of a skilled in the related field, but rather requires the number of data and distinct data to produce good classification conclusions. Convolutional Neural Network (CNN) is used for image classification as well as recognized because of its immense accuracy. In this paper, a comparison between two models of CNN of our selected paper is shown to find the best model to classify tumors in Brain MRI Image and at last, another approach of a CNN model is trained and gained a prediction accuracy of up to 94%. I. INTRODUCTION Medical imaging invokes a number of techniques that can be used as non-interfering methods of looking inside the body. Medical image compasses various image modalities and converts to image of the human body for analyzing and investigating purposes and thus it plays a great and decisive role in taking actions for the enhancement of people’s health. Image segmentation is an essential stride in image process- ing which actuates the accomplishment of a higher level of image processing. The fundamental goal of image segmenta- tion in medical image processing is mainly detecting tumors, competent machine vision and gaining satisfactory result for further diagnosis. Brain, as well as other nervous system cancer, is the 10th leading reason of death and the five-year endurance rate for people with a cancerous brain is 34 percent for men on the other hand 36 percent for women. The World Health Organization (WHO) states that around 400,000 people around the world are affected by the brain tumor and 120,000 people have died in the recent years. Early detection of brain tumors has played an imperative role in developing the treatment possibilities, and a higher gain of survival possibility can be achieved. Although manual segmentation of tumors is a tedious, challenging, and difficult task as it requires a large number of MRI images that are generated in medical routine. Magnetic Resonance Imaging is mainly used for brain tumor detection. Brain tumor seg- mentation from MRI is one of the most compelling tasks in medical image processing as it involves a considerable amount of data. Furthermore, the tumors can be ill-defined with soft tissue boundaries. As a result, it is a very comprehensive task Fig. 1. Dummy Input and Output of Proposed System to gain the accurate segmentation of tumors from the human brain. II. RELATED WORKS In [1] an achievement of substantial results in image segmentation and classification is shown through the convolutional neural network (CNN). A new CNN architecture for brain tumor classification network is simpler than already-existing pre-trained networks, and it was tested on contrast-enhanced magnetic resonance images. In [2] they have established the whole segmentation process based on Mathematical Morphological Operations and applied spatial FCM algorithm which improves the computation time. In [3] they have established that Convolutional Neural Networks are good enough to diagnose brain tumors on MRI images. This study resulted in accuracy of 93% and a loss value of 0.23264. The amount of convolution layers that affects the quality of classification, more convolution layers rise the accuracy results, but more convolution layers will require more time for training. III. OBJECTIVE Convolutional Neural Networks are widely used in the medical field for image processing. Over the years lots after trying researchers built a model which can detect the tumor more accurately. In this paper, the same idea has come up which can accurately identify the tumor from Brain MRI images. A fully connected neural network layer can detect the tumor, but due to parameter sharing and sparsity of connection, CNN is used as our model. The dummy figure of CNN is shown in Figure 1.
  • 2. Fig. 2. Proposed Method A. Convolution Layer Convolution Layer is the main layer in the CNN method which targets to extract features from the input. Convolution executes linear transformations of input data outwardly chang- ing spatial information in the data. Convolution kernels are regulated from the weight of the layer in case the convolution kernels can process the input data training on CNN. B. Subsampling Layer Subsampling focuses on reducing the size of image data and enhancing the invariance of feature positions. CNN applies Max Pooling as a subsampling method. The way Max Pooling works is to break the output of the convolution layer into several smaller grids and after that take the maximum value from each grid to construct a smaller image matrix. Small image size will make it easier to process the next convolution layer C. Fully Connected Layer The Fully Connected Layer evolves the dimensions of the data so that it can be classified linearly. In the convolution layer, each neuron must be converted into one-dimensional data before including into another layer which is connected as a whole. This process is induced by data losing its spatial information and at last Fully Connected Layer network is applied. IV. METHODOLOGY A. Proposed Method This paper implements CNN for the detection of brain tumors. This study accepts input images labeled as yes or no from the raw dataset and then applies these patterns to categorize between tissues that do not contain tumors and those that contain tumors. This paper has implemented 2 model for CNN. An extra model of CNN is also implemented afterwards. Therefore, the proposed system is illustrated in Figure 2. B. Data Augmentation The amount of data in the dataset is not sufficient to be used as training data for CNN. As a result, the augmentation method is used to overcome the imbalance of issues. Augmentation is an algorithm that can utilize statistical data information and form an integrated model. Fig. 3. First CNN Model Fig. 4. Second CNN Model C. Image Pre-processing Pre-processing is executed to create smooth training as there are different variants of intensity, contrast, and size in images. The proportion of the images are resized to shape (240, 240, 3) = (imagewidth, imageheight, numberofchannels) because the images in the dataset have different sizes. After normalization, scale pixel values to the range 0-1 to facilitate the learning process. D. Model CNN In this research, the CNN model contains several layers, namely the convolution layer, the pooling layer, the dense layer, the flatten layer, and the dropout layer. Along with the layers used in the CNN process, there is also an activation function in this study using Relu activation. In this paper, 3 CNN models are made as comparison material. The CNN model design can be seen in Figure 3, Figure 4 and Figure 5. An image in the form of a total association with the first convolution, image size of 240x240 pixels. Kernels have a size of 3x3 and filters are used as many as 32. After that, the model will perform the activation and pooling data functions. The Pooling layer process handles to diminish the dimensions of the feature map. The derivation of the convolution process is mainly a feature map that is employed for the consecutive convolution process repeatedly. The next step is a flatten feature map in vector form to carry out a fully connected layer to produce a classification of images. V. EXPERIMENTS A. Dataset The dataset used in this study is Brain MRI Images for Brain Tumor Detection obtained from kaggle.com. The dataset consists of 253 images assembled into 2 groups, 155 brain
  • 3. Fig. 5. Third CNN Model Fig. 6. First Image without having any Tumor and Second Image having a Brain Tumor images that have tumors, and the rest 98 brain images that do not have tumors. After data augmentation, the dataset subsists of 1085 samples containing tumors and 980 samples not containing tumors, bringing a total of 2065 images. The images having tumor and no tumor is shown in Figure 6. The dataset is splitted into training, testing and validation set with 70:15:15 ratio. B. Evaluation Metrices The evaluation matrices for the three model are repre- sented by Classification Report(Accuracy, Precision, Recall, F1-score), Confusion Matrix as well as Loss Curve. 1) CNN Model 1: For Model 1 the evaluation matrices are shown in Figure 7. Figure 8, Figure 9. Figure 10, Figure 11. Figure 12, Figure 13. Figure 14, Figure 15. Figure 16, Figure 17. Figure 18, Figure 19. Figure 20, Figure 21. Figure 22, Figure 23. Figure 24, Figure 25. Figure 26. Fig. 7. Confusion Matrix and Loss Curve of Model 1 (Run 1) Fig. 8. Classification Report of Model 1 (Run 1) Fig. 9. Confusion Matrix and Loss Curve of Model 1 (Run 2) Fig. 10. Classification Report of Model 1 (Run 2) Fig. 11. Confusion Matrix and Loss Curve of Model 1 (Run 3) Fig. 12. Classification Report of Model 1 (Run 3) Fig. 13. Confusion Matrix and Loss Curve of Model 1 (Run 4) Fig. 14. Classification Report of Model 1 (Run 4) Fig. 15. Confusion Matrix and Loss Curve of Model 1 (Run 5)
  • 4. Fig. 16. Classification Report of Model 1 (Run 5) Fig. 17. Confusion Matrix and Loss Curve of Model 1 (Run 6) Fig. 18. Classification Report of Model 1 (Run 6) Fig. 19. Confusion Matrix and Loss Curve of Model 1 (Run 7) Fig. 20. Classification Report of Model 1 (Run 7) Fig. 21. Confusion Matrix and Loss Curve of Model 1 (Run 8) 2) CNN Model 2: For Model 2, the evaluation matrices are shown in Figure 27. Figure 28, Figure 29. Figure 30, Figure 31. Figure 32, Figure 33. Figure 34, Figure 35. Figure 36, Figure 37. Figure 38, Figure 39. Figure 40, Figure 41. Figure 42, Figure 43. Figure 44, Figure 45. Figure 46. 3) CNN Model 3: For Model 2, the evaluation matrices are shown in Figure 47. Figure 48, Figure 49. Figure 50, Figure Fig. 22. Classification Report of Model 1 (Run 8) Fig. 23. Confusion Matrix and Loss Curve of Model 1 (Run 9) Fig. 24. Classification Report of Model 1 (Run 9) Fig. 25. Confusion Matrix and Loss Curve of Model 1 (Run 10) Fig. 26. Classification Report of Model 1 (Run 10) Fig. 27. Confusion Matrix and Loss Curve of Model 2 (Run 1) Fig. 28. Classification Report of Model 2 (Run 1)
  • 5. Fig. 29. Confusion Matrix and Loss Curve of Model 2 (Run 2) Fig. 30. Classification Report of Model 2 (Run 2) Fig. 31. Confusion Matrix and Loss Curve of Model 2 (Run 3) Fig. 32. Classification Report of Model 2 (Run 3) Fig. 33. Confusion Matrix and Loss Curve of Model 2 (Run 4) Fig. 34. Classification Report of Model 2 (Run 4) Fig. 35. Confusion Matrix and Loss Curve of Model 2 (Run 5) Fig. 36. Classification Report of Model 2 (Run 5) Fig. 37. Confusion Matrix and Loss Curve of Model 2 (Run 6) Fig. 38. Classification Report of Model 2 (Run 6) Fig. 39. Confusion Matrix and Loss Curve of Model 2 (Run 7) Fig. 40. Classification Report of Model 2 (Run 7) Fig. 41. Confusion Matrix and Loss Curve of Model 2 (Run 8) Fig. 42. Classification Report of Model 2 (Run 8)
  • 6. Fig. 43. Confusion Matrix and Loss Curve of Model 2 (Run 9) Fig. 44. Classification Report of Model 2 (Run 9) Fig. 45. Confusion Matrix and Loss Curve of Model 2 (Run 10) Fig. 46. Classification Report of Model 2 (Run 10) 51. Figure 52, Figure 53. Figure 54, Figure 55. Figure 56, Figure 57. Figure 58, Figure 59. Figure 60, Figure 61. Figure 62, Figure 63. Figure 64, Figure 65. Figure 66. Fig. 47. Confusion Matrix and Loss Curve of Model 3 (Run 1) Fig. 48. Classification Report of Model 3 (Run 1) Fig. 49. Confusion Matrix and Loss Curve of Model 3 (Run 2) Fig. 50. Classification Report of Model 3 (Run 2) Fig. 51. Confusion Matrix and Loss Curve of Model 3 (Run 3) Fig. 52. Classification Report of Model 3 (Run 3) Fig. 53. Confusion Matrix and Loss Curve of Model 3 (Run 4) Fig. 54. Classification Report of Model 3 (Run 4) Fig. 55. Confusion Matrix and Loss Curve of Model 3 (Run 5)
  • 7. Fig. 56. Classification Report of Model 3 (Run 5) Fig. 57. Confusion Matrix and Loss Curve of Model 3 (Run 6) Fig. 58. Classification Report of Model 3 (Run 6) Fig. 59. Confusion Matrix and Loss Curve of Model 3 (Run 7) Fig. 60. Classification Report of Model 3 (Run 7) Fig. 61. Confusion Matrix and Loss Curve of Model 3 (Run 8) C. Results Experiments in this article are carried out on 2065 images consisting of 1085 samples containing tumors and 980 samples containing no tumors. The data has been run for 10 times, each using the CNN model that has been made before, each experiment using 25 epochs and 32 batches. Fig. 62. Classification Report of Model 3 (Run 8) Fig. 63. Confusion Matrix and Loss Curve of Model 3 (Run 9) Fig. 64. Classification Report of Model 3 (Run 9) Fig. 65. Confusion Matrix and Loss Curve of Model 3 (Run 10) Fig. 66. Classification Report of Model 3 (Run 10) 1) CNN Model 1: This model gives highest training accu- racy value of 100% and has lowest loss value of 0.002 on the training data, but there is a significant difference with the test data results, the highest test data accuracy value is 83% and the value of F1-Score is 0.825 presented in Table I. 2) CNN Model 2: This model gives highest training accu- racy value of 99.73% and lowest loss value of 0.0.0148, then the test data obtained a highest accuracy value of 90% and F1-Score value of 0.9 presented in Table II. 3) CNN Model 3: This model gives highest accuracy value of 95.64% and lowest loss value of 0.1101, then the test data obtained a highest accuracy value of 94% and F1-Score value of 0.94 presented in Table III.
  • 8. TABLE I RESULT OF FIRST CNN MODEL No Train Validation Test Loss Accuracy Loss Accuracy Accuracy F1-score 1 .0033 1.0 0.9676 .822 0.83 0.825 2 .0076 1.0 0.9521 .7879 0.78 0.78 3 .0042 1.0 0.8097 0.8068 0.8 0.8 4 0.0387 0.996 0.725 0.7689 0.79 0.79 5 0.0094 1.0 0.8585 0.803 0.82 0.815 6 0.0062 1.0 0.8314 0.8068 0.8 0.8 7 0.0042 1.0 0.8243 0.7727 0.8 0.8 8 0.004 1.0 0.8259 0.8106 0.79 0.79 9 0.004 1.0 0.7821 0.7992 0.79 0.795 10 0.002 1.0 0.9391 0.7879 0.8 0.8 Average .00836 0.9996 0.8516 0.7966 0.8 0.7995 TABLE II RESULT OF SECOND CNN MODEL No Train Validation Test Loss Accuracy Loss Accuracy Accuracy F1-score 1 0.0179 0.9953 0.6348 0.8447 0.89 0.885 2 0.202 0.9933 0.3601 0.9129 0.87 0.87 3 0.0267 0.992 0.4388 0.8712 0.88 0.885 4 0.0438 0.9879 0.3646 0.9053 0.89 0.89 5 0.0517 0.9785 0.3507 0.8788 0.89 0.885 6 0.0793 0.9705 0.3205 0.9091 0.89 0.89 7 0.0379 0.9873 0.3942 0.8939 0.88 0.88 8 0.022 0.9933 0.3473 0.9129 0.89 0.89 9 0.0148 0.9973 0.359 0.875 0.88 0.88 10 0.0249 0.9906 0.3892 0.8939 0.9 0.9 Average 0.03392 0.9886 0.396 0.8898 0.886 0.8855 TABLE III RESULT OF THIRD CNN MODEL No Train Validation Test Loss Accuracy Loss Accuracy Accuracy F1-score 1 0.1171 0.9564 0.2091 0.9242 0.94 0.93 2 0.1627 0.9430 0.2242 0.9015 0.94 0.935 3 0.1342 0.9517 0.2203 0.9318 0.94 0.94 4 0.1772 0.9316 0.2982 0.8864 0.9031 0.905 5 0.1485 0.9416 0.3182 0.8750 0.91 0.91 6 0.1326 0.9484 0.2013 0.9318 0.9354 0.93 7 0.1623 0.9437 0.221 0.9318 0.93 0.93 8 0.1453 0.9437 0.2316 0.9053 0.94 0.935 9 0.1319 0.9504 0.2377 0.8977 0.9161 0.915 10 0.1101 0.9544 0.2105 0.9242 0.9129 0.91 Average 0.1422 0.9465 0.2372 0.911 0.9268 0.924 VI. CONCLUSION AND FUTURE DIRECTIONS Convolutional Neural Networks are great enough to diag- nose brain tumors using MRI images. This study resulted in an accuracy of 94% . The count of convolution layers affects the quality of classification shown in Figure 67 as more convolution layers expand the accuracy of results. The process of image augmentation can develop the alternatives of existing datasets, thereby raising the classification results. As this paper is developed by only using CNN, in future, this paper will be developed by using other hybrid deep learning algorithm. Fig. 67. Comparison of Models REFERENCES [1] Milica M Badža and Marko Č Barjaktarović. Classification of brain tumors from mri images using a convolutional neural network. Applied Sciences, 10(6):1999, 2020. [2] B Devkota, Abeer Alsadoon, PWC Prasad, AK Singh, and A Elchouemi. Image segmentation for early stage brain tumor detection using mathemat- ical morphological reconstruction. Procedia Computer Science, 125:115– 123, 2018. [3] DC Febrianto, I Soesanti, and HA Nugroho. Convolutional neural network for brain tumor detection. In IOP Conference Series: Materials Science and Engineering, volume 771, page 012031. IOP Publishing, 2020.