A Hybrid System Based on FMM and MLP to Diagnose Heart Disease

In the medical field diagnosis of a disease at an early stage is very important. Nowadays soft computing techniques such as fuzzy logic, artificial neural network and Neuro- fuzzy networks are widely used for the diagnosis of various diseases at different levels. In this chapter, a hybrid neural network is designed to classify the heart disease data set the hybrid neural network consist of two types of neural network multilayer perceptron (MLP) and fuzzy min max (FMM) neural network arranged in a hierarchical manner. The hybrid system is designed for the dataset which contain the combination of continuous and non continuous attribute values. In the system the attributes with continuous values are classified using the FMM neural networks and attributes with non-continuous value are classified by using the MLP neural network and to synthesize the result the output of both the network is fed into the second MLP neural network to generate the final result.

On the Implementation of a Digital Image Watermarking Framework Using Saliency and Phase Congruency

In this paper, a saliency and phase congruency based digital image watermarking scheme has been projected. The planned technique implants data at least significant bits (LSBs) by means of adaptive replacement. Here more information is embedded into less perceptive areas within the original image determined by a combination of spectral residual saliency map and phase congruency map. The position of pixels with less perceptibility denotes the most unimportant region for data hiding from the point of visibility within an image. Therefore any modification within these regions will be less perceptible to one observer. The model gives a concept of the areas which has excellent data hiding capacity within an image. Superiority of the algorithm is tested through imperceptibility, robustness, along with data hiding capacity.

For Better Healthcare Mining Health Data

Data mining has great contributions to the healthcare such as support for effective treatment, healthcare management, customer relation management, fraud and abuse detection and decision making. The common data mining methods used in healthcare are Artificial Neural Network, Decision trees, Genetic Algorithms, Nearest neighbor method, Logistic regression, Fuzzy logic, Fuzzy based Neural Networks, Bayesian Networks and Support Vector Machines. The most used task is classification. Because of the complexity and toughness of medical domain, data mining is not an easy task to accomplish. In addition, privacy and security of patient data is a big issue to deal with because of the sensitivity of healthcare data. There exist additional serious challenges. This chapter is a descriptive study aimed to provide an acquaintance to data mining and its usage and applications in healthcare domain. The use of Data mining in healthcare informatics and challenges will be examined.

Adaptive Edge Detection Method towards Features Extraction from Diverse Medical Imaging Technologies

The CAD is a relatively young interdisciplinary technology, has had a tremendous impact on medical diagnosis specifically cancer detection. The accuracy of CAD to detect abnormalities on medical image analysis requires a robust segmentation algorithm. To achieve accurate segmentation, an efficient edge-detection algorithm is essential. Medical images like USG, X-Ray, CT and MRI exhibit diverse image characteristics but are essentially collection of intensity variations from which specific abnormalities are needed to be isolated. In this chapter a robust medical image enhancement and edge detection algorithm is proposed, using tree-based adaptive thresholding technique. It has been compared with different classical edge-detection techniques using one sample two tail t-test to exam whether the null hypothesis can be supported. The proposed edge-detection algorithm showing 0.07 p-values and 2.411 t-stat where a = 0.025. Moreover the proposed edge is single pixeled and connected which is very significant for medical edge detection.

A Combinational Fuzzy Clustering Approach for Microarray Spot Segmentation

Due to microarray experiment imperfection, spots with various artifacts are often found in microarray image. A more rigorous spot recognition approach in ensuring successful image analysis is crucial. In this paper, a novel hybrid algorithm was proposed. A wavelet approach was applied, along with an intensity-based shape detection simultaneously to locate the contour of the microarray spots. The proposed algorithm segmented all the imperfect spots accurately. Performance assessment with the classical methods, i.e., the fixed circle, adaptive circle, adaptive shape and histogram segmentation showed that the proposed hybrid approach outperformed these methods.

Dealing with Higher Dimensionality and Outliers in Content-Based Image Retrieval

This chapter proposes a content based image retrieval method dealing with higher dimensional feature of images. The kernel principal component analysis (KPCA) is done on MPEG-7 Color Structure Descriptor (CSD) (64-bins) to compute low-dimensional nonlinear-subspace. Also the Partitioning Around Medoids (PAM) algorithm is used to squeeze search space again where the number of clusters are counted by optimum average silhouette width. To refine these clusters further, the outliers from query image's belonging cluster are excluded by Support Vector Clus-tering (SVC). Then One-Class Support Vector Machine (OCSVM) is used for the prediction of relevant images from query image's belonging cluster and the initial retrieval results based on the similarity measurement is feed to OCSVM for training. Images are ranked from the positively labeled images. This method gives more than 95% precision before recall reaches at 0.5 for conceptually meaningful query categories. Also comparative results are obtained from: 1) MPEG-7 CSD features directly and 2) other dimensionality reduction techniques.

Data Clustering and Various Clustering Approaches

It is needed to organize the data in different groups for various purposes, where clustering is useful. The chapter covers Data Clustering in the detail, which includes; introduction to data clustering with figures, data clustering process, basic classification of clustering and applications of clustering, describing hard partition clustering and fuzzy clustering. Some most commonly used clustering method are explained in the chapter with their features, advantages, and disadvantages. A various variant of K-Means and extension method of hierarchical clustering method, density-based clustering method and grid-based clustering method are covered.

Proficient Clustering algorithm for Wireless Sensor Networks

Enhancing the energy efficiency and maximizing the networking lifetime are the major challenges in Wireless Sensor Networks (WSN).Swarm Intelligence based algorithms are very efficient in solving nonlinear design problems with real-world applications.In this paper a Swarm based Fruit Fly Optimization Algorithm (FFOA) with the concept of K-Medoid clustering and swapping is implemented to increase the energy efficiency and lifetime of WSN. A comparative analysis is performed in terms of cluster compactness,cluster error and convergence. MATLAB Simulation results show that K-Medoid Swapping and Bunching Fruit Fly optimization (KMSB-FFOA) outperforms FFOA and K-Medoid Fruit Fly Optimization Algorithm (KM-FFOA).

Graph Theoretic Approaches for Image Analysis

Different graph theoretic approaches are prevalent in the field of image analysis. Graphs provide a natural representation of image pixels exploring their pairwise interactions among themselves. Graph theoretic approaches have been used for problem like image segmentation, object representation, matching for different kinds of data. In this chapter, we mainly aim at highlighting the applicability of graph clustering techniques for the purpose of image segmentation. We describe different spectral clustering techniques, minimum spanning tree based data clustering, Markov Random Field (MRF) model for image segmentation in this respect.

Association Rules-Based Analysis in Multidimensional Clusters

Many approaches for identifying potentially interesting items exploiting commonly used techniques of multidimensional data analysis. There is a great need for designing association-rule mining algorithms that will be scalable not only with the number of records (number of rows) in a cluster but also among domain's size (number of dimensions) in a cluster to focus on the domains. Where the items belong to domain is correlated with each other in a way that the domain is clustered into classes with a maximum intra-class similarity and a minimum inter-class similarity. This property can help to significantly used to prune the search space to perform efficient association-rule mining. For finding the hidden correlation in the obtained clusters effectively without losing the important relationship in the large database clustering techniques can be followed by association rule mining to provide better evaluated clusters.