Convolution Neural Network with Unsupervised Machine Learning Approach for Feature Extraction and Brain Tumor Detection in Human beings

Brain tumors are atypical progress of cells in the brain or the contiguous tissues. Tumors can be either non-cancerous or cancerous. The prognosis for a person with a brain tumor varies significantly. Benign tumors are generally less aggressive and may be curable with surgery. Malignant tumors can be more challenging to treat and may have a poorer prognosis. The outcome also depends on the stage at which the tumor is diagnosed. Detecting brain tumors using magnetic resonance imaging (MRI) images with convolutional neural networks (CNNs) is a common and effective approach in medical image analysis. In this research, the unsupervised machine learning approach called self-organizing map (SOM) is implemented for effective feature extraction and CNN with ResNet architecture is employed for brain tumor exposure efficiently. The proposed SOMResNet algorithm takes the MRI images as the input and perform feature extraction. Again, the mined essential features are given as the input for SOMResNet to identify the tumor in the human brain. The accuracy of the proposed SOMResNet is compared with softmax, ReLu, Tanh, decision tree classifiers. The accuracy of SOMResNet is 97.5%, Sensitivity is 98.0%, specificity rate is 97.1%, precision rate is 97.3% with 97.6% of F-Score value. The result shows that the SOMResNet algorithm outperforms than the traditional algorithms in brain tumor detection.