Madhubanti Maitra

A Slantlet transform based intelligent system for magnetic resonance brain image classification

Abstract The present paper proposes the development of a new approach for automated diagnosis, based on classification of magnetic resonance (MR) human brain images. Wavelet transform based methods are a well-known tool for extracting frequency space information from non-stationary signals. In this paper, the proposed method employs an improved version of orthogonal discrete wavelet transform (DWT) for feature extraction, called Slantlet transform, which can especially be useful to provide improved time localization with simultaneous achievement of shorter supports for the filters. For each two-dimensional MR image, we have computed its intensity histogram and Slantlet transform has been applied on this histogram signal. Then a feature vector, for each image, is created by considering the magnitudes of Slantlet transform outputs corresponding to six spatial positions, chosen according to a specific logic. The features hence derived are used to train a neural network based binary classifier, which can automatically infer whether the image is that of a normal brain or a pathological brain, suffering from Alzheimers disease. An excellent classification ratio of 100% could be achieved for a set of benchmark MR brain images, which was significantly better than the results reported in a very recent research work employing wavelet transform, neural networks and support vector machines.

Gene selection from microarray gene expression data for classification of cancer subgroups employing PSO and adaptive K-nearest neighborhood technique

A PSO-adaptive KNN based gene selection method is proposed to select useful genes.A heuristic for selecting the optimal values of K efficiently is also proposed.The proposed technique is applied on SRBCT, ALL_AML and MLL microarray datasets.The usefulness of the identified genes is reconfirmed using SVM classifier.The method finds 6, 3 and 4 genes for SRBCT, ALL_AML, and MLL with high accuracy. These days, microarray gene expression data are playing an essential role in cancer classifications. However, due to the availability of small number of effective samples compared to the large number of genes in microarray data, many computational methods have failed to identify a small subset of important genes. Therefore, it is a challenging task to identify small number of disease-specific significant genes related for precise diagnosis of cancer sub classes. In this paper, particle swarm optimization (PSO) method along with adaptive K-nearest neighborhood (KNN) based gene selection technique are proposed to distinguish a small subset of useful genes that are sufficient for the desired classification purpose. A proper value of K would help to form the appropriate numbers of neighborhood to be explored and hence to classify the dataset accurately. Thus, a heuristic for selecting the optimal values of K efficiently, guided by the classification accuracy is also proposed. This proposed technique of finding minimum possible meaningful set of genes is applied on three benchmark microarray datasets, namely the small round blue cell tumor (SRBCT) data, the acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML) data and the mixed-lineage leukemia (MLL) data. Results demonstrate the usefulness of the proposed method in terms of classification accuracy on blind test samples, number of informative genes and computing time. Further, the usefulness and universal characteristics of the identified genes are reconfirmed by using different classifiers, such as support vector machine (SVM).

Nonlinear state feedback controller design for underactuated mechanical system: A modified block backstepping approach

This paper presents the formulation of a novel block-backstepping based control algorithm to address the stabilization problem for a generalized nonlinear underactuated mechanical system. For the convenience of compact design, first, the state model of the underactuated system has been converted into the block-strict feedback form. Next, we have incorporated backstepping control action to derive the expression of the control input for the generic nonlinear underactuated system. The proposed block backstepping technique has further been enriched by incorporating an integral action additionally for enhancing the steady state performance of the overall system. Asymptotic stability of the overall system has been analyzed using Lyapunov stability criteria. Subsequently, the stability of the zero dynamics has also been analyzed to ensure the global asymptotic stability of the entire nonlinear system at its desired equilibrium point. The proposed control algorithm has been applied for the stabilization of a benchmarked underactuated mechanical system to verify the effectiveness of the proposed control law in real-time environment.

Robust medical image segmentation using particle swarm optimization aided level set based global fitting energy active contour approach

The active contour models have been popularly employed for image segmentation for almost a decade now. Among these active contour models, the level set based Chan and Vese algorithm is a popular region-based model that inherently utilizes intensity homogeneity in each region under consideration. However, the Chan and Vese model often suffers from the possibility of getting trapped in a local minimum, if the contour is not properly initialized. This problem assumes greater importance in the context of medical images where the intensity variations may assume large varieties of local and global profiles. In this work we propose a robust version of the Chan and Vese algorithm which is expected to achieve satisfactory segmentation performance, irrespective of the initial choice of the contour. This work formulates the fitting energy minimization problem to be solved using a metaheuristic optimization algorithm and makes a successful implementation of our algorithm using particle swarm optimization (PSO) technique. Our algorithm has been developed for two-phase level set implementation of the Chan and Vese model and it has been successfully utilized for both scalar-valued and vector-valued images. Extensive experimentations utilizing different varieties of medical images demonstrate how our proposed method could significantly improve upon the quality of segmentation performance achieved by Chan and Vese algorithm with varied initializations of contours.

Classification of overcurrent and inrush current for power system reliability using Slantlet transform and artificial neural network

If a fault occurs in the distribution system, for protectivity and reliability, the fault should be immediately cleared and the system should be re-energized. It is mandatory to ensure that upon re-energization, overcurrent, if encountered, may be due to inrush of current and not for the persistent fault. Hence, it could be impossible to protect the distribution system unless the fault current can be distinguished from the inrush current. Different measures are to be taken to handle the two different events of fault and inrush. Hence, the authors pose the problem as a classification problem, where, the fault current can be deterministically separated from inrush current. The concept of classification, in general, depends on some characteristic features of the events, which are the key components in which the events differ. For automated classification, these distinguishing features of the events are to be judiciously extracted first. This work proposes a novel scheme for automated feature extraction, using Slantlet Transform (ST) and subsequently an automated classification mechanism based on Artificial Neural Network (ANN). ST has been regarded as a contemporary development in the field of multiresolution analysis, which is proposed as an improvement over the discrete wavelet transform (DWT). For each candidate inrush or fault current waveform, suitable features are extracted by employing ST. Then, a successfully trained ANN based classifier, developed utilizing inputs comprising the features extracted from a training set of waveforms, is implemented for a testing set of sample waveforms. The proposed scheme could achieve 100% classification accuracy in the testing phase.

A novel scheme for feature extraction and classification of magnetic resonance brain images based on Slantlet Transform and Support Vector Machine

Automated diagnosis of various brain abnormalcies is possible if classification of magnetic resonance (MR) human brain images can be carried out in an efficacious manner. The present paper proposes the development of a new approach for automated diagnosis, which rests on classification of brain magnetic resonance imaging (MRI) techniques. In our present work we propose a method that uses an improved version of orthogonal discrete wavelet transform (DWT) for feature extraction, called Slantlet transform, which can especially be useful to provide superior time localization with simultaneous achievement of shorter supports for the filters. The features, hence, obtained are used to train a support vector machine (SVM) based binary classifier that automatically infers whether the images that of a normal brain or that of a pathological one. An excellent classification ratio of 100% could be achieved for a set of benchmark MR brain images, which is significantly better than the results reported in a recent research work employing combination of different feature extraction and classification tools e.g. wavelet transform, neural networks and SVM.

An Intelligent Paging Strategy Using Rule-Based AI Technique for Locating Mobile Terminals in Cellular Wireless Networks

This paper proposes an efficient rule-based paging strategy (RBPS) using a well-known concept of artificial intelligence, namely, rule base. The novelty of the scheme lies in devising ldquorulesrdquo that offer a potential mapping from seemingly disparate input data items (yet having some statistical relations) to an almost exact position of mobile terminals (MTs). Considering the conventional models of call arrival, cell residence, and mobility, we have developed a stochastic model to analyze the performance of the scheme. Interestingly, RBPS requires no additional processing at MTs and involves a nominal overhead at mobile switching centers. Simulation results reveal that RBPS significantly outperforms the blanket paging scheme adopted in global system for mobile (GSM) communications. In addition, results are very much encouraging when compared with the popular shortest-distance-first scheme. Finally, RBPS is generic enough to be potentially used in next-generation wireless networks, irrespective of any standards, with only minor adaptations to conform to the respective standards.

An intelligent location management strategy for next generation wireless networks: a rule base approach

The paper proposes an integrated location management scheme, based on our previously proposed novel rule-based paging scheme (RBPS) for reducing the cost of terminal paging in a cellular wireless environment. We extend that scheme to propose a complete location management scheme. This requires application of RBPS together with a judicious movement based location update (LU) scheme. Numerical results show that the proposed integrated location management scheme, called RBPS*, not only attempts to resolve the issue of inherent tradeoff between the cost components, comprised of paging and LU cost, but also does reduce the total cost for location management compared to the other schemes. RBPS* significantly outperforms RBPS and the GSM-adopted blanket paging scheme along with movement based LU. When compared with the popular shortest distance first (SDF) scheme, results are again encouraging. The proposed scheme, RBPS*, is a generic one and can be deployed on-line. It has the potential for use in next generation wireless networks, irrespective of any standards.

Autonomous robot coverage in rescue operation

Mobile robots are integrated into a search and rescue team as tools for searching victims in dangerous areas that is harmful for human, so as to follow the human entity during the mission. Generalized Local Voronoi Diagram (GLVD) algorithm is introduced to model the area, coverage and cooperation. The senor based motion planning is implemented on the robots for navigation of the area and coverage. The focus here is on the rescue of entities present following area coverage. A brief description of the on-going research and the results obtained is also provided.

A heuristic-based approach for minimization of inter-switch handoff cost with load balancing for optimal location area planning in a PCSN

Our work is a variant of the location area planning (LAP) problem where location areas (LA) are formed by assigning microcells of the total coverage area of a personal communication services network (PCSN) to mobile switching centers (MSC) or switches dedicated to that network. The criterion is minimization of total signaling load imposed upon the system due to the generation of handoff requests by the mobile terminals (MT) in the system and initiation of the procedure for performing successful handoff towards completion at a positive rate. The cell-to-switch assignment has been done in such a way that no switches are overloaded. Moreover, we show that loads on the switches are almost balanced even if multiple boundary crossings by an MT requests several handoffs for a single call.