Image segmentation is still an important and interesting issue for researchers today, and one of MRI issues is image segmentation of brain tumors. MRI images of brain tumors have quite high complexity both from variable size and structure, low contrast, bad boundaries, and low homogeneity. In some cases noise appears like blur, miss focus, and miss alignment of images. Manually image segmentation requires high accuracy, since each pixel contains important information. In addition, it will potentially lead to different interpretations of each expert. The tools that are used by experts to read MRI images of the brain nowadays, can present images into several regions namely sagittal, coronal, and cross sectional. The division into these three regions is actually quite helpful during the diagnosis process, since the doctor can see the brain tissue in the three regions, but the display cannot lead to an object that is found slightly different from particular area or focus on one particular object or area that shows a difference or abnormality. The expected outcome of this initial research is an architecture design for segmenting MRI images of brain tumors by involving two algorithms, MDWT and CNN. Combination both of MDWT and CNN provides new approaches in medical images analysis including brain image tumor segmentation. Researchers use literature study and laboratory experiment as the research methods.
- Convolution neural network
- Image segmentation
- Magnetic resonance image
- Multi directional wavelet transform