Automatic Brain Tumor Detection in Medical Imaging using Machine Learning

The segmentation of brain tumor is a very crucial task for detecting the tumor in the early stage. Many methods of brain tumor segmentation are available in the literature but none of the method achieved satisfactory performance. Brain tumor has caused more deaths compared to other cases related to cancer. Due to the complex structure of the brain tumor and its merging with the normal tissues, the classification of tumor from normal tissues is a challenging task. In this article, we have proposed LIPC based methodology for brain tumor classification and segmentation. The main idea is to take the segmentation challenge as a multiclass classification task. MRI sequences of the brain have common characteristics (locality) and are sparse (sparsity). We have utilized both these properties in our proposed methodology. Firstly the image preprocessing techniques are used for noise removal and image enhancement. After that different textural features were calculated and reduced by PCA for better classification score. The dataset used is that of MICCAI 2013 containing High-grade tumor and Low-grade tumor for real patient data. Our method presented a Dice Score of 0.95 for complete tumor which is by far compared to the methods already available in terms of Dice Score and time complexity.