Brain tumors are a group of cancers that originate from different cells of the central nervous system or cancers of other tissues in the brain. Excessive cell growth in the brain is called a tumor. Tumor cells need food and blood to survive. Growth and proliferation of tumor cells in the cranial space, cause strain inside the brain and thus disrupt vital human structures. Therefore, diagnosis in the early stages of brain tumors is crucial. This study introduces a new optimized method for early diagnosis of the brain tumor. The method has five main parts of noise reduction, tumor segmentation, morphology, feature extraction based on wavelet and gray-level co-occurrence matrix, and classification based on an optimized deep belief network. For optimizing the classifier network, an enhanced version of the moth search algorithm is utilized. Simulation results are applied to three different datasets, FLAIR, T1, and T2, and the accuracy results of the presented method are compared with two other metaheuristics, particle swarm optimization and Bat algorithms. The final results showed that the presented technique has good achievements toward the compared methods.
Latent Correlation Representation Learning for Brain Tumor Segmentation with Missing MRI Modalities
