Political exponential deer hunting optimization-based deep learning for brain tumor classification using MRI

In this work, three deep learning classifiers, namely pyramid scene parsing network (PSPNet), Shepard convolutional neural network (ShCNN) and deep convolutional neural network (Deep CNN), are introduced to segment, detect and classify the tumor region from magnetic resonance images (MRI). The pre-processing is carried out by exploiting region of interest extraction that removes the irrelevant noise that exists in the images. The PSPNet segments the tumor region from the MRI to progress the efficacy of classification. The ShCNN detects the segmented image as either normal or abnormal image. Moreover, the tumor classification is completed using Deep CNN, which classifies the detected image as either benign, edema, core or malignant. In addition, the training process of three classifiers is performed by exploiting the adopted political exponential deer hunting optimization algorithm, which is designed by incorporating political optimizer, and exponential deer hunting optimization. Moreover, the performance analysis proved that the adopted tumor detection method obtained the optimal performance based on the testing accuracy, sensitivity, as well as specificity of 0.904, 0.919 and 0.899, and the developed classification approach obtained maximum testing accuracy, sensitivity, as well as specificity of 0.929, 0.935 and 0.902, respectively.

Automated summary

In this study, three deep learning classifiers, namely PSPNet, ShCNN, and Deep CNN, are used to segment, detect, and classify tumor regions from MRI. Pre-processing is done by extracting the region of interest to remove irrelevant noise. PSPNet segments the tumor regions for better classification. ShCNN detects the segmented images as normal or abnormal. Deep CNN is used for tumor classification into benign, edema, core, or malignant. The classifiers are trained using a political exponential deer hunting optimization algorithm. Performance analysis shows high accuracy, sensitivity, and specificity for tumor detection and classification.