B-Scopus

Permanent URI for this communityhttps://dspace.psgrkcw.com/handle/123456789/3730

Browse

Has files: YesSubject: search.filters.subject.Deep learning

Search Results

Now showing 1 - 7 of 7
  • No Thumbnail Available
    Item
    ARTIFICIAL INTELLIGENT MODELS FOR AUTOMATIC DIAGNOSIS OF FOETAL CARDIAC ANOMALIES: A META-ANALYSIS
    (Springer Link, 2023-01-01) Divya, M O; Vijaya, M S
    The foetal anomaly scanning is one of the most challenging areas where accuracy of diagnosis much fluctuating with respect to the expertise of the radiologist and the mental equilibrium of the radiologist at the time of scanning. Amongst the various anomalies, foetal heart anomaly diagnosis expects precise and sensitive intellectual presence since perilous congenital heart diseases are one of the common causes resulting in the major population of infant mortality or into permanent natal faults. The accuracy of manual diagnosis of foetal cardiac abnormalities from the ultrasound scan images vary based on the human expertise and the presence of mind. Therefore, the scope of computer-assisted judgement can produce accurate diagnosis irrespective of the operator’s profile. Numerous researches are going on to explore the scope of computer-assisted judgement of abnormalities using ultrasound imaging technique (USIT), specifically using machine learning and deep learning models. This work exploits the opportunities of computer-assisted diagnosis in foetal cardiac anomaly diagnosis as this is one of the most sensitive areas where appropriate diagnosis can save a life and a wrong diagnosis may lose a life unnecessarily.
  • No Thumbnail Available
    Item
    OPTIMIZING PRE-PROCESSING FOR FOETAL CARDIAC ULTRA SOUND IMAGE CLASSIFICATION
    (Springer Link, 2023-03-28) Divya, M O; Vijaya, M S
    Recent research shows that Foetal cardiac anomalies which gets diagnosed postnatally makes a grave negative impact on the delivery outcome. The situation becomes lethal when severe anomalies get diagnosed after the baby is born. Many medical researches shows that delivery outcome could be better when the anomaly is diagnosed prenatally. There are hardly any research and development happening in this area where automation and prediction are on prime focus for finding the cardiac anomaly using Ultra Sound Imaging Technique (USIT). The USIT during the second trimester is universal for every pregnant woman also the second trimester is the best time to take appropriate medical assistance for the foetus in case of anomaly. This research is experimental study to setup a standard dataset for foetal cardiac anomaly USITs and to identify the appropriate pre-processing technique for binary classification of USIT. The 1200 images in the dataset are organised in two classes half of the images are with anomaly and other half without anomaly. The class with anomaly includes images representations from 17 anomalies which is theoretically established as structural anomalies of heart. All anomalies are present in the dataset approximately equal in ratio. The dataset has undergone the following pre-processing techniques, blur removal, noise removal and contrast normalisation. The Alex-net model is trained to create a binary classifier for the FetalEcho dataset after applying the different pre-processing techniques. Eight rounds of classification have been performed with eight versions of the FetalEcho dataset. The worst results were shown by the row dataset (FetalEcho_V01) when the classification experiment have been performed with the AlexNet classifier. The dataset FetalEcho_V05, created after removing blur and noise, is identified as the best performance for classification, amongst the eight datasets.
  • Thumbnail Image
    Item
    DEEP POSITIONAL ATTENTION-BASED HIERARCHICAL BIDIRECTIONAL RNN WITH CNNBASED VIDEO DESCRIPTORS FOR HUMAN ACTION RECOGNITION
    (International Journal of Intelligent Engineering and Systems, 2022-03-18) Srilakshmi, Nagarathinam; Radha, Narayanan
    Human Action Recognition (HAR) is a highly notable area of study in contemporary computer vision. Many investigations focused on recognizing a person’s actions from video streams based on extracting features regarding orientation and motion. This article presents a Joints and Trajectory-pooled 3D-Deep Positional Attention (PA)-based Hierarchical Bidirectional Recurrent Convolutional Descriptors (JTDPAHBRD) approach which uses a PA-based Hierarchical Bidirectional Recurrent Neural Network (PAHBRNN) for enhancing the feature aggregation process. First, the entire video is segregated into multiple blocks and they are provided to the 2-stream bilinear Convolutional 3D (C3D) model which applies the PAHBRNN as feature aggregation. In PAHBRNN, the feature vectors related to the different parts of a human skeleton in a certain clip are hierarchically aggregated using the position-aware guidance vector. Then, 2 different streams in the C3D network are fused and trained end-to-end using the softmax loss to get the final video descriptor for a particular video sequence. Further, the Support Vector Machine (SVM) classifier is applied to classify the resultant video descriptor to recognize the person’s actions. At last, the investigational outcomes demonstrate the JTDPAHBRD achieves 99.6% better recognition accuracy than the classical state-of-the-art approaches
  • No Thumbnail Available
    Item
    BUILDING ACOUSTIC MODEL FOR PHONEME RECOGNITION USING PSO-DBN
    (Inderscience Enterprises Ltd, 2020-04-02) Laxmi Sree, B R; Vijaya, M S
    Deep neural networks has shown its power in generous classification problems including speech recognition. This paper proposes to enhance the power of deep belief network (DBN) further by pre-training the neural network using particle swarm optimisation (PSO). The objective of this work is to build an efficient acoustic model with deep belief networks for phoneme recognition with much better computational complexity. The result of using PSO for pre-training the network drastically reduces the training time of DBN and also decreases the phoneme error rate (PER) of the acoustic model built to classify the phonemes. Three variations of PSO namely, the basic PSO, second generation PSO (SGPSO) and the new model PSO (NMPSO) are applied in pre-training the DBN to analyse their performance on phoneme classification. It is observed that the basic PSO is performing comparably better to other PSOs considered in this work, most of the time.
  • No Thumbnail Available
    Item
    DEEP LEARNING PREDICTIVE MODEL FOR DETECTING HUMAN INFLUENZA VIRUS THROUGH BIOLOGICAL SEQUENCES
    (Springer Link, 2020-09-08) Nandhini, M; Vijaya, M S
    Swine influenza is a contagious disease which is generated by one of the swine influenza viruses. Any modification in protein will alter the biological activity and lead to illness. Obtaining appropriate information from virus protein sequence is an interesting research problem in bioinformatics. The aim of this research work is to develop deep neural network (DNN)-based virus identification model for detecting the virus accurately with the protein sequences using deep learning. Deep learning is gaining more importance because of its governance in terms of accuracy when the network trained with large amount of data. A corpus of 404 protein sequences associated with nine types of human influenza virus is collected for training the deep neural network and building the model. Various parameters of the DNN such as input layer, hidden layer and output layer are fine-tuned to improve the efficiency of the model. Sequential model is created for developing DNN classification model using Adam optimizer with Softmax and ReLu activation functions. It is observed that experiments of proposed human influenza virus identification model with DNN classifier give 80% of accuracy and outperform with other ensemble learning algorithms.
  • No Thumbnail Available
    Item
    DEEP NEURAL NETWORK FOR EVALUATING WEB CONTENT CREDIBILITY USING KERAS SEQUENTIAL MODEL
    (Springer Link, 2020-08-08) Manjula, R; Vijaya, M S
    Web content credibility determines the measure of acceptable and reliable of the web content that is observed. Content will prove to be unreliable if it is not updated, and it is not controlled for remarkable, and therefore, web content credibility is considerably essential for the people to assess the content. The analysis of content credibility is a vital and challenging task as the content credibility is outlined on crucial factors. This paper focuses on building deep neural network (DNN)-based predictive model using sequential model API to evaluate credibility of a webpage content. Deep neural network (DNN) is considered as an extremely promising decision-making architecture, and it performs feature extraction and transformation with the use of refined statistical modeling. A corpus of 400 webpage contents has been developed, and the factors like readability, freshness, and duplicate content are defined and captured from the webpage content. These features are redefined, and a new set of features is self-learned through the deep layers of neural network. Numeric labeling is used for defining credibility, wherein five-point Likert scale rating is used to denote the content credibility. By using sequential model, KerasRegressor with ADAM optimizer and a multilayer network is generated for building DNN-based predictive model and discovered that deep neural network outperforms other general regression algorithms in prediction scores.
  • No Thumbnail Available
    Item
    A REVIEW ON CLASSIFICATION AND RETRIEVAL OF BIOMEDICAL IMAGES USING ARTIFICIAL INTELLIGENCE
    (Springer Link, 2021-08-12) Greeshma, K V; Viji Gripsy, J
    Image retrieval and classification are the most prominent area of research in computer vision. Nowadays, bounteous medical images are generated through different types of medical imaging modalities in healthcare systems. It is often very difficult for researchers and doctors to access manage and retrieve images easily. The efficient and effective analysis and usage of heterogeneous biomedical images growing rapidly are a tedious task. Content-based image retrieval (CBIR) is one of the most widely used methods for automatic retrieval of images and widely used in medical images. Abundant research articles are published in different domain of applications related to CBIR and classification. The aim of this study is to provide a road map for researchers by exploring the various approaches, techniques, and algorithms used for medical image retrieval and classification. Feature extraction is the main subject for improving the performance of image classification and retrieval. Bag of visual words techniques and deep convolutional neural networks are widely used in content-based medical image retrieval (CBMIR). The state-of-the-art methods presented in this review are well suited to classify and retrieve multimodal medical images for different body organs. The methods include preprocessing of images, feature extraction, classification, and retrieval steps to develop an efficient biomedical image retrieval system. This chapter briefly reviews the various techniques used for biomedical images, and different methods adopted in classification and retrieval are focused.