Skip to Main Content
This research addresses to the problem of establishment patterns, training of the optimal desired network, classification and accuracy prediction. Once the accuracy issue had been addressed, the issue of efficiency detection and maintenance is normally introduced and experimented in most of the scenarios. The later stage of development is dealt with the inter racial classification of diseases which gets over lapped. The problem of detection of brain images into corresponding diseased and Non-Diseased syndrome types for functional Magnetic Resonance Images. The classification problem gets intensified when there comes the inter lapped diseased contents. Among such, overlapped problems occur when there are significant inter-related similar symptoms, similar types of traditional symptoms, common symptoms, where there needs to be minute differences in disease identification. A case occurs with relevance to the shrinking of the frontal lobes and expansion of the back lobes. When a Tumor or a cyst starts protruding, pain occurs followed blurred vision and nausea, unbearable pain and others symptoms. Some of the Alzheimer's symptoms are the shrinking of the frontal lobes and expansion of the behind lobes, followed by memory loss, abnormal behaviour and etc., The process involves the classification among Tumour Vs Alzheimers' classification. The images differ with capturing intensity and noise content. This happens during motion capturing. The choosing up of the images for featuring detection, feature vector calculation and subjecting to training and testing is all carried out from fMRI images. The reason being the functional magnetic Resonance Image properties gets inducted in the fMRI images compared to the other types of images like the PET, CT images etc., A Daubechies Wavelet Transform acting as a smoothening filter was applied playing a dual role in denoising, Decomposition and extraction of frequency feature components for feature vector calculation. A modified med- an filter removes random noises better which has been induced. A fourth level order of Daubechies Discrete Wavelet decomposition was used for feature vector formation. This was performed because of the blurring results of the images due to the wavelet implication. This naturally destroy the feature coefficients but yet another prominent and effective way in order the classification of the brain images into diseased and non - diseased brain images. The approximated sub - band images were chosen from the input images to train the network. In designing a network using a competitive neural network as a subclass to classify the normal and abnormal image type into corresponding diseased and non - disease types. Initially subjecting to a Competitive Learning, Self - Organized Maps are enhanced to Self - Organized Feature Maps and finally enhanced to Linear Vector Quantization. On comparing to the earlier competitive network, the network of Competitive Learning, SOM and SOFM, and invoking Linear Vector Quantization were considered as improvised for classification and efficient for detection.