B. Rajesh Kanna

Bio: B. Rajesh Kanna is an academic researcher from VIT University. The author has contributed to research in topics: Digital image & Pixel. The author has an hindex of 4, co-authored 21 publications receiving 71 citations. Previous affiliations of B. Rajesh Kanna include St. Joseph's College of Engineering & Shanmugha Arts, Science, Technology & Research Academy.

Tuberculosis (TB) detection system using deep neural networks

Abstract: Microscopy is a rapid diagnosis method for many infectious diseases like tuberculosis (TB). In TB bacilli identification, specimens are stained using Ziehl–Neelsen or Auramine dye and are examined by technicians thoroughly for any infectious microbes. For pathological study, the images of these microbes are captured using microscopes and image processing is applied for further analysis. However, choosing 100 field of views (FOV) randomly from a 2 × 1 cm square area of sputum specimen may lead to inconsistency in specificity. The examination of specimens is a tedious process, and it requires especially skilled technicians for screening the sputum smear samples. The proposed tuberculosis detection system consists of two subsystems—a data acquisition system and a recognition system. In the data acquisition system, a motorized microscopic stage is designed and developed to automate the acquisition of all FOVs. Here the microscopic stage movement is motorized and scanning patterns are defined by the user for specimen examination. After the acquisition of all FOVs, data are passed to the recognition system. In the recognition system, transfer learning method is implemented by customizing the Inception V3 DeepNet model. This model learns from the pre-trained weights of Inception V3 and classifies the data using support vector machine (SVM) from the transferred knowledge. For training and testing the customized Inception V3 model, a public TB dataset (Shah et al. in J Med Imaging 4(2):027503, 2017. https://doi.org/10.1117/1.jmi.4.2.027503 ) and our own acquired microscopic digital dataset are used for analysis. In this model, the fixed feature representations are taken from the top stack layer of Inception V3 DeepNet and are classified using SVM. This model attains an accuracy of 95.05%, thereby reducing the dependency on skilled technicians in the screening process and increasing sensitivity and specificity.

ColorFingers: improved multi-touch color picker

Abstract: ColorFingers is a WYSIWYG, Location Independent Touch (LIT) based color picking tool aimed to give unique and swift interaction in choosing color on touch based devices. It makes use of touch interface and the touch information of two fingers to select almost 16 million colors. This tool is a model to prove how touch can be interpreted in different ways to achieve performance improvements in HCI. In this paper, we propose ColorFingers which is a color picker and briefly discuss the working of it. We show, how it achieves around 54% reduction in color selection time and 53% improvement in accuracy when compared to existing models. The proposed model emphasizes on Multi-touch, Quick feedback and Location Independency.

Image-based area estimation of any connected region using y-convex region decomposition

Abstract: In this paper, we address the issue of finding the area of any connected region from its digital image. We have reviewed the existing techniques and brought out their shortcomings. In particular, we observe that none of the existing techniques can satisfactorily find the area of certain types of regions. To overcome such shortcomings, we propose an algorithm for decomposing any connected region R into one or more disjoint regions Y i . The decomposition is such that all the sub-regions satisfy a convexity property, referred to as y -convex property. y -Convex region is one in which no vertical line intersects the boundary curve more than twice. An interesting property of a y -convex region is that its contour can be split into an upper curve and a lower curve, and hence known numerical methods can be used to estimate its area. So, our decomposition algorithm enables us to find the area of R as the summation of the areas of all Y i . Since there are no bench mark images for validation, we have created a new database of images including the images used by previous works. Experimental results show that our method has an average accuracy of 99.01% and the paired t -test confirms that the difference between the actual and estimated area is considered to be not quite statistically significant with t and two-tailed p values of 1.68 and 0.0953, respectively. The proposed approach can be potentially used in applications such as geographical survey, morphological analysis of micro-organisms, medical image analysis, crime scene documentation, agriculture, and food industry.

Development of yConvex hypergraph model for contour-based image analysis

Abstract: To recognize and classify any connected region-of-interest (ROI) in a digital image, the shape descriptor of the ROI has to be defined in terms of its boundary characteristics. However, for certain types of regions, it is very difficult to exactly trace the contour and none of the existing techniques satisfactorily work. In this paper, yConvex hypergraph model (yCHG) of digital image is introduced. It is a generalization of yConvex region decomposition [1] and represents any connected region as a finite set of disjoint yConvex hyperedges (yCHE). Each yCHE is a yConvex region, whose contour can be tracked in a deterministic way. Based on this hypergraph model (yCHG), an algorithm is presented for contour-based shape representation. Representations of contours of individual yCHEs are obtained independently and combined to provide a representation for the whole shape. This representation is rotation and scale invariant. The proposed algorithm works correctly on any connected region including those that can not be satisfactorily handled by any of the existing techniques.

An efficient intrusion detection system based on hypergraph - Genetic algorithm for parameter optimization and feature selection in support vector machine

Abstract: Realization of the importance for advanced tool and techniques to secure the network infrastructure from the security risks has led to the development of many machine learning based intrusion detection techniques. However, the benefits and limitations of these techniques make the development of an efficient Intrusion Detection System (IDS), an open challenge. This paper presents an adaptive, and a robust intrusion detection technique using Hypergraph based Genetic Algorithm (HG - GA) for parameter setting and feature selection in Support Vector Machine (SVM). Hyper – clique property of Hypergraph was exploited for the generation of initial population to fasten the search for the optimal solution and to prevent the trap at the local minima. HG-GA uses a weighted objective function to maintain the trade-off between maximizing the detection rate and minimizing the false alarm rate, along with the optimal number of features. The performance of HG-GA SVM was evaluated using NSL-KDD intrusion dataset under two scenarios (i) All features and (ii) informative features obtained from HG – GA. Experimental results show the prominence of HG-GA SVM over the existing techniques in terms of classifier accuracy, detection rate, false alarm rate, and runtime analysis.

Deep learning approach to detect malaria from microscopic images

Abstract: Malaria is an infectious disease which is caused by plasmodium parasite. Several image processing and machine learning based techniques have been employed to diagnose malaria, using its spatial features extracted from microscopic images. In this work, a novel deep neural network model is introduced for identifying infected falciparum malaria parasite using transfer learning approach. This proposed transfer learning approach can be achieved by unifying existing Visual Geometry Group (VGG) network and Support Vector Machine (SVM). Implementation of this unification is carried out by using “Train top layers and freeze out rest of the layers” strategy. Here, the pre-trained VGG facilitates the role of expert learning model and SVM as domain specific learning model. Initial ‘k’ layers of pre-trained VGG are retained and (n-k) layers are replaced with SVM. To evaluate the proposed VGG-SVM model, a malaria digital corpus has been generated by acquiring blood smear images of infected and non-infected malaria patients and compared with state-of-the-art Convolutional Neural Network (CNN) models. Malaria digital corpus images were used to analyse the performance of VGG19-SVM, resulting in classification accuracy of 93.1% in identification of infected falciparum malaria. Unification of VGG19-SVM shows superiority over the existing CNN models in all performance indicators such as accuracy, sensitivity, specificity, precision and F-Score. The obtained result shows the potential of transfer learning in the field of medical image analysis, especially malaria diagnosis.

Detection of tuberculosis from chest X-ray images: Boosting the performance with vision transformer and transfer learning

Abstract: Tuberculosis (TB) caused by Mycobacterium tuberculosis is a contagious disease which is among the top deadly diseases in the world. Research in Medical Imaging has been done to provide doctors with techniques and tools to early detect, monitor and diagnose the disease using Artificial Intelligence. Recently, many attempts have been made to automatically recognize TB from chest X-ray (CXR) images. Still, while the obtained performance is encouraging, according to our investigation, many of the existing approaches have been evaluated on small and undiverse datasets. We suppose that such a good performance might not hold for heterogeneous data sources, which originate from real world scenarios. Our present work aims to fill the gap and improve the prediction performance on larger datasets. In particular, we present a practical solution for the detection of tuberculosis from CXR images, making use of cutting-edge Machine Learning and Computer Vision algorithms. We conceptualize a framework by adopting three recent deep neural networks as the main classification engines, namely modified EfficientNet, modified original Vision Transformer, and modified Hybrid EfficientNet with Vision Transformer. Moreover, we also empower the learning process with various augmentation techniques. We evaluated the proposed approach using a large dataset which has been curated by merging various public datasets. The resulting dataset has been split into training, validation, and testing sets which account for 80%, 10%, and 10% of the original dataset, respectively. To further study our proposed approach, we compared it with two state-of-the-art systems. The obtained results are encouraging: the maximum accuracy of 97.72% with AUC of 100% is achieved with ViT_Base_EfficientNet_B1_224. The experimental results demonstrate that our conceived tool outperforms the considered baselines with respect to different quality metrics.