IMPROVED BRAIN PATHOLOGY CLASSIFICATION USING A HYBRID DEEP LEARNING ALGORITHM: A NOVEL APPROACH

Abstract

The most significant obstacle to treatment in neurology is early tumor diagnosis. Because brain tumors are so frequent, there is a lot of research into ways to spot cancer early. Automating and diagnosing using traditional image processing methods is difficult. Radiologists and doctors now have a new tool at their disposal to help diagnose brain tumors more quickly and with more confidence. This technique makes use of CNNs. Research and synthesis are aided in distinguishing benign from malignant occurrences by the increased feature maps generated by the proposed deep learning architecture. Two DNNs are combined in the suggested H-DNN design. Two methods have been developed based on information extracted from magnetic resonance imaging (MRI) scans; one uses spatial texture data from cranial images, and the other uses frequency domain data. In the end, we combine the two neural networks to make prediction score-based classification even better. In contrast to DNN-1, which makes use of Local Binary Patterns for training, DNN-2 makes use of frequencies from Wavelet Transformation. Both the Real MRI dataset and the BraTS T2-weighted MRI dataset were used to test the proposed model. The model used in this investigation achieved the best classification accuracy, at 98.7 percent, according to related work. The reported model fared better than both the DNN-1 and DNN-2 designs when comparing the accuracy, sensitivity, and specificity of the proposed technique.