Brain Tumor Glioma Analysis Through Computational Intelligence

Abstract

The field of image processing offers distinctive features and is useful in medical diagnostics and imaging system. For the radiologists, manually identifying and classifying the Tumor has become a demanding and frantic process. Brain Magnetic resonance (MR) images must be extracted from malignant Tumor areas, which is a laborious and time-consuming task carried out by radiology experts or healthcare professionals. Current studies now heavily rely on medical imaging mainly to the continuous progress in automated brain Tumor classification and segmentation. This aids in quick decision as well as clear vision, diagnosis, and easier medication progression for the professionals. A dynamically Deep Learning technique for Glioblastoma brain cancer survival prediction rate was put out to address the aforementioned problems. In this research thesis, we present two approaches for detecting the brain tumor, risk prediction and measure the survival rate of patient. In the first approach, we developed the computer-aided tumor diagnosis techniques based on CNN that have demonstrated to be effective and have contributed considerable strides in computer vision. The deep learning method for predicting the prognosis of glioma brain tumors is covered in this research. Glioma prediction has been determined using MRI brain tumor imaging. Data pre-processing is the initial phase. The MRI brain images were improved by intensities normalization using histogram normalization, de-noising via bilateral filtering, and the removal of information contaminants. Probabilistic noise salted and peppers distortion was also taken out. Secondly, radiomic features segmentation was completed using the MFCM clustering approach. Then, Rough Set Theory-based Grey Wolf Optimization was used to choose the most important and instructive aspects from the obtained characteristics. Then, using FR- CNN, the overall survival predictions categorization is performed to the important feature selection in MRI brain images. The proposed MFCM-RSGWO-FRCNN approach is tested against state-of-arts FCM, OTSUS, NB, and SVM approaches. Evaluation parameters like Specificity, Sensitivity, PSNR, Mean Square Error (MSE), Segmentation Time, and Prediction Accuracy were used to examine the technique. The proposed MFCM-RSGWO-FRCNN has the advantages of less converging and the corresponding characteristics. In the second approach, machine learning technique of the Random Forest model and Deep Neural Network method is proposed to predict the glioblastoma recurrence risk. Initially, Resampling and Z-Score Normalization are the image preprocessing techniques that are used to remove the outlier in MRI brain image data. After the pre-processing brain images are then segmented using the Recurrent Neural Network-Generative Adversarial Network (RNN-GAN), which mitigates the impact of imbalanced pixel labels. Subsequently, the Wavelet Band-Pass Filtering technique is presented to extract the texture features and the CE-T1WI model predicts PFS and ORR in recurrent GBM patients treated with the combination of Nivolumab and Bevacizumab. Accordingly, Random Forest and DNN techniques are proposed for patients’ recurrence risk prediction. After analyzing both models, the second RNN+GAN Model gives a better result as compared to the first MFCM RSGWO-FRCNN Model. The RNN+GAN Model achieves a 95.11% accuracy score; Sensitivity is also 95.11% and Specificity is 98%. The RNN+GAN Model increase the survival rate which is 2.47% after diagnosis and overall treatment of patients. And finally, RNN+GAN Model is compared with existing state-of-art methods.