Feature Extraction And Pattern Recognition Using Soft Computing In Medical Imaging

Abstract

newline newlineFeature Extraction and pattern recognition play a vital role in medical image processing. Therefore, in this work, a more accurate and reliable brain tumor classification method has been proposed for designing CAD systems by using soft computing. A publicly available dataset contains 3064 MRI images have been used for training the classifiers. Biorthogonal filters have proven to be effective in comparison to orthogonal wavelet filters. The study also depicts that the biorth 3.9 wavelet filter is the most effective filter among all the wavelet filters. newline The second step of the preprocessing stage is skull removal and normalizing the background. Morphological opening and closing operations with FCM algorithm have been used for segmenting the tumor. From this tumor region, total 14 features form three categories named shape, intensity and texture, are extracted using GLCM and GLDM techniques. To choose the best feature vectors from the trained feature matrices, three optimization methods, Grey Wolf Optimization , Particle Swarm Optimization and Firefly Algorithm have been applied. newline From the segmented output, it is necessary for the classifier to identify the type of brain tumor based on features extracted from a segmented region because the treatment of every type of tumor is not the same. The results show that the performance of the SVM classifier is better than kNN and Naive Bayes. The performance of the proposed Hybrid framework named GWO SVM achieves the accuracy for Meningioma tumor is 99.30 percent, for Glioma tumor is 97.67 percent and for Type 3 Pituitary tumor is 98.23 percent. So the mean accuracy for the proposed GWO SVM model is 98.40 percent approx. Other parameters like sensitivity 99.15 percent, specificity 99.66 percent and balanced accuracy 99.40 percent are also showing promising results with the GWO SVM model. newline newline newline