Lizy Abraham

Bio: Lizy Abraham is an academic researcher from APJ Abdul Kalam Technological University. The author has contributed to research in topics: Deep learning & Phonocardiogram. The author has co-authored 2 publications.

Digital image forensic using deep flower pollination with adaptive Harris hawk optimization

Abstract: This paper presents a new segmentation algorithm and deep learning concept for innovative copy-move image forgery identification. In this work, the segmentation algorithm uses a new Adaptive Harris Hawk Optimization (AHHO) algorithm. The host image is segmented into irregular and non-overlapping blocks, named as Image Blocks (IB). Here the segmented image is compressed with a Discrete Cosine Transform (DCT). After that, the Zernike moments and Gabor filter-based features extraction techniques are processed to extract the Block Features (BF). At last, the tampered portions are classified with the hybrid Deep Neural Network (DNN) and Flower Pollination Algorithm (FPA) for forgery classification. This novel deep learning algorithm reduces the learning complexities and improves detection accuracy. The compression attacks are removed with anti-forensic blocking artefact removal concept. The experiments are conducted on the Benchmark dataset, CoMoFoD, and GRIP. For the benchmark dataset, precision, recall and F1 values are 91.27,100.0 and 96.07, respectively. For CoMoFoD dataset, the values of precision, recall and F1 are 92.57, 98.0 and 93.05, respectively. For GRIP dataset, the values of precision, recall and F1 are 97.02, 98.0 and 93.05, respectively. The implementation outcomes demonstrated that the proposed scheme is most effective in copy-move forgery recognition than the existing forgery detection approaches.

Detection and Classification of Cardiovascular Disease from Phonocardiogram using Deep Learning Models

Abstract: Cardiovascular disease (CVD) is one of the prime reason for death in India and across the globe. Rural areas of India suffer from shortage of cardiologist and medical facilities. Hence there is a need for the development of an efficient, automated heart disease detection system that can analyse the phonocardiogram to detect the disease. The paper proposes deep learning architectures for anomaly detection from heart sounds. The work classifies the unsegmented phonocardiograms into five classes, four cardiovascular diseases and normal(N). The detected pathological conditions are mitral valve prolapse (MVP), mitral stenosis (MS), mitral regurgitation (MR) and aortic stenosis (AS). Features are extracted using Mel Frequency Cepstral Coefficient (MFCCs) and learning and classification are performed using deep learning methods such as Convolutional Neural Network (CNN), Long Short Term Memory (LSTM) and a combination of 1DCNN and LSTM. A total of 1960 phonocardiogram (PCG) segments are used to develop the models with 392 segments in each class. We have achieved an accuracy of 99.1%, 98.2%, 99.4% for CNN, LSTM and 1DCNN-LSTM respectively.

A Novel Analysis of Performance and Inference Time of Machine Learning Models to Detect Cardiovascular Emergency Situations of Rescue Patients

Abstract: Cardiovascular complications are considered as one of the most common and fatal complications to the rescue personnel and require urgent medical intervention to save an emergency patient. Often due to the delay in detecting heart complication in urgent situation, necessary treatment paths cannot be followed which results in high mortality rate. Although the current researches show promising aspects in detecting cardiovascular diseases in early stage, they are focused on detecting only some of specific and common cardiovascular diseases from clinically recorded or long term historical patients’ data. The novel approach followed in this research is: instead of using traditional health data collected in clinical environment, we developed the model with 9 years of rescue mission’s real-time recorded data to recognise any cardiovascular situation in general. To find out the best model, different machine learning(ML) algorithms like Support Vector Machine(SVM), Random Forest(RF), K-nearest neighbour(KNN), Extreme Gradient Boosting(XGB), Logistic Regression(LR), Naive Bayes(NB) and Artificial Neural Network(ANN) were used. From the performance comparison, we concluded that extreme gradient boosting and neural network showed the best performance in terms of all evaluation parameters. Fast inference is the basic requirement for any rescue mission. So an inference time analysis of the ML models and Apache-TVM machine learning compiler was shown to understand their compatibility in real world applications.