Automated brain tumor segmentation from mri data based on exploration of histogram characteristics of the cancerous hemisphere
| Autor: | Samee Azad, S.M. Khan, Shaikh Anowarul Fattah, Mir Khadiza Akter |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
020205 medical informatics
medicine.diagnostic_test Pixel business.industry Computer science Brain tumor Magnetic resonance imaging Pattern recognition 02 engineering and technology Image segmentation medicine.disease Histogram Brain size 0202 electrical engineering electronic engineering information engineering medicine Median filter Artificial intelligence business Classifier (UML) |
| Zdroj: | 2017 IEEE Region 10 Humanitarian Technology Conference (R10-HTC). |
| DOI: | 10.1109/r10-htc.2017.8289080 |
| Popis: | Accurate detection of a brain tumor from 3D MRI images is very important for the physicians to provide proper treatment to the patients diagnosed with fatal diseases. If the detection of the tumor region is done manually, it is a very prolonged task to analyze a single case. Often it is erroneous as well. This can create adverse effect on planning the treatment of the patient. Therefore, in this work, a completely automated method of detection of the brain tumor has been proposed. Human brain size being huge, and often the characteristics of tumor tissues and non-tumor tissues having similarity, it is very difficult and time consuming for a classifier to work with the entire brain data. This paper deals with detecting the brain tissue so accurately that the classifier will require only a very small volume to work on. This method takes out 2D slices of images from the 3D data and then detects the tumor by investigation of the features derived from the histograms of the slices. At first, 2D slices have been taken along the XY plane and the tumorous hemisphere is detected from the intensity histogram of the two hemispheres. A threshold intensity is determined by analyzing the histogram of the detected hemisphere. After applying the threshold, median filtering is performed and a second threshold value is applied if needed. After that, a connectivity checking is performed on the image and the biggest cluster is selected as pixels representing the tumor. Finally, the 2D slices containing the detected tumor are stacked upon and unified together. The proposed method, with Dice Similarity Coefficient Metric of 0.8056, has surpassed many other algorithms. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|