Automatic System for Detection and Classification of Brain Tumors
- 1. Welcome our Graduation project presentation July 2016
- 2. Minia University Faculty of Engineering Biomedical Engineering Department By: Basma Adham, Enas Leashaa, Fatma Sayed, Heba Abdel-Razic,Walid Salah Supervisor: Dr. Ashraf Mahroos
- 4. Outlines Abstract Motivation Introduction Methodology Comparative study Results Conclusion Future Work
- 5. Abstract • Automatic system for brain tumors detection based on DICOM MRI images • Surveying methodologies of from preprocessing to classifications • Implementing comparative study. • Proposed technique with highest accuracy and lest elapsed time.
- 6. Motivation of the project Prevalence of human brain tumors Why this Project Project challenges State of art of Automatic Brain Tumor Detection Proposed techniques
- 7. Prevalence of brain tumors • The National Brain Tumor Foundation (NBTF) for research in United States estimates that 29,000 people in the U.S are diagnosed with primary brain tumors each year. • Population of Egypt in 2012.... 83.9 million • People diagnosis with cancer 108,600 • Risk of getting cancer before age 75: 15.4% • People dying from cancer /year : 72,300
- 8. • Software diagnostic application expensive and not widely used in Egypt, so we help doctors in our region to get right decision. • This project helps medical staff in diagnosis which is the first and main part of treatment of any disease. • The project not widely made in Egypt so we made a start in this kind of research in upper Egypt. Why this Project
- 9. Why this Project • Brain cancer remains one of the most incurable forms of cancer, with an average survival period of one to two years. • It is not easy to deal with a tumor in it like the rest of the body's organs. • There are more than 120 types of brain tumors.
- 10. • The brain is divided into regions that control various functions. • Damage to a region may affect the functions it controls. Why this Project
- 11. Dataset availability was the biggest problem faced us Skull has the same intensity of tumor. large elapsed time in some techniques. Accuracy of classifications. Project challenges
- 12. State-of-the-art of Automatic Brain Tumor Detection • 2006 was the first to use Digital awvelet transform (DWT ) coefficients to detect pathological brains. Projects are developed and search progress till 2016.
- 14. Introduction Brain anatomy Types of brain tumors Proposed flow chart
- 15. Brain anatomy Normal brain anatomy parts. The brain is composed of three parts: the brainstem, cerebellum, and cerebrum. The cerebrum is divided into four lobes: frontal, parietal, temporal, and occipital
- 16. Brain tumors malignant Astrocytomas Glioblastoma. benign Brain tumors malignant primary secondary benign Types of brain tumors
- 17. Data Acquisition MRI principles Data availability problem
- 18. Magnetic Resonance Imaging (MRI) • The main diagnostic tools of brain tumor are CT and MRI. • MRI provides a much greater contrast between the different soft tissues of the body than (CT) does.
- 21. The MRI of the brain can be divided into three regions: • white matter (WM) • gray matter (GM) • cerebrospinal fluid (CSF)
- 22. Dataset availability • Dataset of patients in hospitals cannot be provided without security approval. • The need of DICOM format • large variety of brain tumors • Data descriptions Main dataset is obtained from Safwa Radiology Center. Second dataset from The Cancer Imaging Archive
- 23. Preprocessing applied techniques Normalization Skull stripping Median Filter Gaussian Filter Histogram Equalization Edge detection Morphological operation preprocessing Skull stripping Median Filter Gaussian Filter Histogram Equalization Thresholding Normalization Contrast Enhancement Anisotropic diffusion Filter Edge detection
- 24. Why normalization. 𝑰 = (𝑰−𝑰_𝒎𝒊𝒏) (𝑰𝒎𝒂𝒙−𝑰𝒎𝒊𝒏) Datasets Normalization Preprocessing
- 26. Median Filter The median filter is a nonlinear digital filtering technique, often used to remove salt and paper noise Preprocessing
- 27. Original image Median filtered image Preprocessing
- 28. Gaussian Filter • Constant factor at front makes volume sum to 1 • Remove “high-frequency” components from the image (low-pass filter) • Smoothing and bluring an image. 0.003 0.013 0.022 0.013 0.003 0.013 0.059 0.097 0.059 0.013 0.022 0.097 0.159 0.097 0.022 0.013 0.059 0.097 0.059 0.013 0.003 0.013 0.022 0.013 0.003 5 x 5, = 1 Preprocessing
- 29. Original image Image after Gaussian filter Preprocessing
- 30. Histogram Equalization • The histogram equalization is an approach to enhance a given image. • Increase the intensity range of pixels • Make smoothing image 3 2 4 5 7 7 8 2 3 1 2 3 5 4 6 6 8 5 11 13 18 18 20 5 5 1 5 8 13 11 15 18 The original image equalized image Preprocessing
- 31. Histogram Equalization algorithm 1. Count the total mo. of pixels with each pixel intensity 2. Calculate probability of each pixel intensity. 3. Probability is no. of pixels divided by total no. of pixels 4. Calculate cumulative probability 5. multiply cumulative probability by constant no. 6. round the decimal no. obtained integer no. Preprocessing
- 32. Brain image with tumor Histogram Equalization Image after Histogram equalization Preprocessing
- 33. Edge detection A location in the image where is a sudden changing in the intensity/color of pixels Preprocessing
- 34. Edge detection Brain image without skull Prewitt filter Preprocessing
- 35. Thresholding Original image Threshold level = o.5 Preprocessing
- 36. Contrast enhancement Contrast enhancement improves the image quality by enhancing hidden information and gives better quality. Enhancing the contrast of images is done by transforming the values in an intensity image, such that the histogram of the output image approximately matches a specified histogram. Original image before contrast enhancement Image after contrast enhancement
- 37. Histogram before contrast enhancement Histogram after contrast enhancement
- 38. Morphological operation Morphological operation divides into four categories : Erosion Dilation Opening Closing
- 41. Segmentation techniques K-means clustering Otsu segmentation Region growing Watershed segmentation Fuzzy C means
- 43. Segmentation techniques K-means clustering Otsu segmentation Region growing Watershed segmentation Fuzzy C means
- 44. Region growing Region growing technique groups pixels which have same properties based on homogeneity criterion.
- 46. Region growing original image After skull removal Segmentation
- 47. Region growing Output of region growing a b complement of region growing Segmentation
- 48. Region growing a b multiplied image Final image after post processing Segmentation
- 49. K-means Clustering Flowchart of K means algorithm Segmentation
- 50. K means segmentation in each iteration. Original image Filtered image. Segmentation
- 51. Segmented tumor after post processing Segmentation
- 53. Otsu segmentation Otsu segmentation a c b Original image Skull removal Gaussian filter Segmentation
- 54. Otsu segmentation Otsu segmentation for four thresholds The output of Otsu Final image after normalization Segmentation
- 55. Watersheld segmentation 1. Make a binary image 2. Take the nearest Euclidean distance 3. Iterate steps. Segmentation
- 57. Features Extraction Gray-Level Co-Occurrence Matrix (GLCM) Gray Level Difference Matrix Feature reduction using PCA Feature Extraction
- 58. Gray-Level Co-Occurrence Matrix (GLCM) GLCM Principle Feature Extraction
- 59. The gray level difference method (GLDM) The gray level difference method (GLDM). For different values of d we calculated five texture features: energy, standard deviation, Mean skewness and kurtosis. Feature Extraction
- 60. Feature reduction using Principle component analysis (PCA) Summarization of data with many (p) variables by a smaller set of (k) derived variables. n p A n k X Feature Reduction
- 61. PCA algorithm 1- Compute the mean of the data matrix. µ = 1 𝑁 𝑖=1 𝑁 𝒳ᵢ 2- Subtract the mean from each image. 3- Compute the covariance matrix. K=WWᵀ 4- Compute the Eigen values λ and Eigen vectors e for covariance matrix. Solve : K e = λ e. 5-Order them by magnitude: λ 1> λ2>… λN The eigenvalue λ measures the variation in the direction Feature Reduction
- 62. Classifications Neural Network K-nearest neighbor (KNN) classifier Support Vector Machine (SVM)
- 63. Artificial Neural Network Biological neural network Classification
- 64. Architecture of a typical artificial neural network Classification
- 66. K-nearest neighbor (KNN) classifier KNN Classification
- 69. GUI using region growing segmentation Results
- 70. GUI using K means segmentation
- 72. Results of segmentation without skull removal
- 77. Conclusion • Finally; With K meand , GLCM feature extraction and back propagation network Algorithms has been successfully tested and achieved the best results with accuracy 96.7%. with 2.2s.
- 78. Future work •Segmentation 3D tumors, But This work need available database for 3D. •Obtaining the boundary of the tumor and plotting the safety margin in 2D and 3D tumor. This will help doctors to make surgery based on automatic boundary calculation. Furthermore, protect patient from further biopsy procedure. • Complete system with EEG to detect abnormalities with high accuracy.
- 79. Thank You