Transactions on Machine Intelligence 2821-1693 2 4 2019 12 01 Brain Tumor Segmentation in MRI Images Using Transform Domain Methods 221 234 220849 10.47176/TMI.2019.221 EN A. AshamiPour Department of Computer Science, Buin Zahra Branch, Islamic Azad University, Buin Zahra, Iran H. Yarahmadi Assistant Professor, Department of Computer Science, Buin Zahra Branch, Islamic Azad University, Buin Zahra, Iran J. Mohammadzadeh Assistant Professor, Department of Computer Science, Buin Zahra Branch, Islamic Azad University, Buin Zahra, Iran Journal Article 2019 08 04 Image segmentation techniques are widely used in medical imaging to isolate homogeneous regions. To date, no complete image segmentation method has been presented that can yield satisfactory results for imaging applications such as brain MRI, brain cancer detection, and others. This study presents a method for brain tumor segmentation in MRI images using transform domain methods and contourlets. First, a multi-resolution representation of the input image is created using the contourlet transform. Then, an 8-dimensional feature vector is extracted for each pixel using inter-resolution and intra-resolution data. The final feature vector's dimensions are reduced using Principal Component Analysis (PCA). Finally, the feature vectors are grouped into discrete clusters for segmentation. The proposed method was implemented on brain images using MATLAB software. The algorithm is computationally simple, yet efficient for brain tumor segmentation in MRI images. Using the eight subbands, as opposed to the conventional wavelet transform that extracts coefficients in only three directions, helps us better identify directional details in the image. Additionally, the use of eight directional features for each pixel allows the extracted details in different subbands to be enhanced in accordance with the direction, making maximal use of the subband correlation. The proposed method does not suffer from the region overlapping problem of active contour methods and can detect the internal areas of large regions. Overall, the proposed algorithm improves performance by six percent compared to the active contour method and by one percent compared to the two-dimensional feature extraction method using wavelet transform. https://www.tmachineintelligence.ir/article_220849_69db1c22b087a2e9ab06db852d93ec05.pdf