Hierarchical based tumor segmentation by detection using deep learning approach

Abstract

Brain tumor is a cluster of abnormal cells that grows out of control in brain. Identifying brain tumor is challenging for doctors, since its impact will lead to danger for human life. Spotting of brain tumor using traditional methods is not accurate. Deep learning provides solution for detecting Brain Tumor in an efficient way. We have used MRI scan images. Since the image contains noise, image pre-processing work has been done to enhance the images. Deep learning methods for images works with Convolutional Neural Network (CNN). CNN has an advantage of extracting features by own. CNN has many hidden layers, where features are extracted and those features are learned for future prediction process. Single Shot Detector is used for detection of tumor region. SSD uses 8732 default bounding boxes mapped to the ground truth boxes for localisation process. Jaccard Overlap is used for match the default box with ground truth box. The detected whole tumor region is then used for segmenting the proper tumor region.