I am grateful to the numerous local and global colleagues who contributed to the completion of this dissertation. First of all, I would like to thank Dr. Renu John for his advice during my dissertation. As my supervisor, he has continually encouraged me to stay focused on achieving my goal.
I would also like to thank Dr Harikrishnan Narayanan Unni and Dr Subha Narayan Rath for their encouragement and guidance.
Background
Motivation
Contribution of the thesis
Organization of the thesis
Introduction
- Nervous system
- Skull
Brain anatomy
- Brain lobes
- Deep structures
- Cranial nerves
- Meninges
- Ventricles and cerebrospinal fluid
- Blood supply
Pituitary gland: It is located at the base of the skull in the sella turcica, which is a small case of bone. It is called the master gland as it controls the other endocrine glands in the body. The brain communicates with the rest of the body through the spinal cord and cranial nerves.
The cerebellum, brainstem, and lower part of the brain are supplied by the vertebral arteries. The basilar artery and the internal carotid artery join in a structure at the base of the brain called the Circle of Willis [5]. If one of the main vessels is blocked, it is possible for collateral blood flow to pass through the Circle of Willis and.
Venous circulation in the brain follows a different mechanism compared to the rest of the body. In other parts of the body, for blood to drain a certain area, arteries and veins must flow together.
Brain cells
- Nerve cells
- Glia cells
In the brain there are no pairs of internal carotid veins or vertebral veins corresponding to the internal carotid arteries and vertebral arteries. These venous sinuses collect blood from the brain and supply it to the internal jugular veins.
- Classification of brain tumours
- Primary brain tumours
- Secondary brain tumours
- Causes
- Symptoms
- Diagnosis
- Imaging tests
- Biopsy
- Treatment
- Observation
- Medication
- Surgery
- Radiation
- Chemotherapy
- Glioma
Tumor cells that look similar to normal cells are called differentiated cells, while those that look different are called anaplastic cells. Secondary brain tumors are caused by tumors from other parts of the body that spread to the brain. Frontal lobe tumors usually cause behavioral and emotional changes, memory loss, paralysis on one side of the body and loss of vision.
Brain stem tumors can cause difficulty speaking and swallowing, drowsiness, hearing loss, muscle weakness on one side of the body, vomiting, etc. Initially, a complete physical examination is performed and the medical history of the patient and the patient's family is reviewed. A neurological examination is performed to check the patient's mental status and memory, nerve function, muscle strength, coordination, reflexes and response to pain.
Computed Tomography (CT): A series of x-ray images are taken from different angles to generate various 2D views of the brain. In a needle biopsy, a small hole is drilled in the skull, and a hollow needle is inserted through it into the tumor to remove a tumor sample. Surgery is preferred if it is possible to reach the tumor without damaging any important brain regions.
If the tumor cannot be completely removed without damaging vital regions of the brain, only part of the tumor is removed. Radiation is delivered so that tumor cells receive the maximum dose and normal cells receive the minimum dose. Radiation from this material will destroy any tumor cells that may remain in the surrounding tissue.
Chemotherapy drugs can be administered orally, intravenously, or in the form of a wafer into the tumor. There are different types of gliomas, depending on the type of glial cells from which they originate.
- Introduction
- Manual and Automated segmentation
- Manual segmentation
- Semi-automatic segmentation
- Automatic segmentation
- Unsupervised and Supervised segmentation
- Unsupervised segmentation
- Supervised segmentation
- Threshold-based segmentation
- Global threshold
- Local threshold
- Region-based segmentation
- Region growing
- Watershed segmentation
- Pixel classification
- Fuzzy C-Means (FCM) clustering
- Markov Random Fields (MRF)
- Artificial Neural Networks (ANN)
- Model-based segmentation
- Parametric deformable model
- Geometric deformable model
When imaging brain tumor, it tries to segment the image into tumor, edema and other regions. For patient-specific training, the training data is taken from the image to be segmented. The choice of training data is very important as different training data lead to different results.
For each pixel, the threshold is calculated using local statistical properties such as the average intensity value [16] of the neighboring pixels. Region-based segmentation divides the image into different regions by merging adjacent pixels with homogeneity properties based on similarity criteria. The different watersheds are the different regions into which the image is segmented, while the watershed lines serve as the boundaries between the regions.
This is in contrast to hard clustering, where each pixel is assigned to only one class. In brain tumor imaging, spatial information is important because if a pixel is strongly labeled as a tumor, its neighboring pixels are likely to be labeled as a tumor as well. In the training phase, the parameters in the mathematical operations are optimized to minimize the prediction error.
This is useful for brain tumor imaging, as brain tumor data may not follow a simple Gaussian distribution. At each iteration step, the contour points are updated to new values. If any part of the contour comes into contact with the required function, that part pulls the rest of the curve towards the required function.
Parametric models do not converge properly if the required feature has concavities on its boundary. It is also important to place the initial contour near the required feature. GVF uses the spatial diffusion of the edge map gradient instead of the image gradient as the external force.
- Data
- Contrast adjustment
- Region growing
- Introduction
- Region characteristics
- Seed selection
- Neighbourhood selection
- Homogeneity criterion
- Termination of region growing
- Results
- Fuzzy C-Means (FCM) clustering
- Introduction
- Algorithm
- Results
- Volume measurements
The algorithm starts from a point inside the region and grows outward until it reaches the boundaries of the region. These definitions of the image areas may vary depending on the application and the type of image being segmented. There is a chance that the selected starting point falls on a pixel that is not characteristic of the region, even though it is within the region.
Any of the points that lie within a certain predefined range of the average is chosen as the starting point. This yields a new structure which consists of the original structure and its nearest neighboring pixels. This criterion is used to determine the similarity of neighboring pixels to the region.
The homogeneity criterion can be based on intensity or any other image property [30]. For each neighboring pixel, the intensity difference is calculated with the average of the current region. This intensity difference threshold depends on the intensity of the pixels in the image.
In any step of the iteration, if none of the adjacent pixels meet the homogeneity criterion, the area does not grow in that step. The locations of the reference points and the assignment of the data points to clusters are adjusted and applied over the images to be segmented. Usually, when a data set is partitioned into clusters, each data point is definitively assigned to one of the clusters.
The partition matrix gives the cluster labels (usually the centers) and the membership function of each pixel in each cluster. Fuzzy membership functions, constrained to lie between 0 and 1, indicate the pixel's similarity to each of the tissue classes. If the initialization is done using a rough estimate of the cluster centers, the algorithm converges faster and the results are more accurate.
From each of the segmented results, the number of pixels representing tumor and edema is counted. Tumor/edema volume is found by multiplying the number of pixels by the volume of a voxel.
Summary
Conclusion
Future work
Sasi Kumar, Skull stripping and automatic segmentation of brain MRI using seed growth and threshold techniques. Segmentation and classification of brain tumors on apparent diffusion coefficient images using self-organizing maps.
