Minal Deshmukh

Bio: Minal Deshmukh is an academic researcher from Vishwakarma Institute of Information Technology. The author has contributed to research in topics: Computer science & Digital image processing. The author has an hindex of 1, co-authored 3 publications receiving 7 citations.

Arrhythmia classification using morphological features and probabilistic neural networks

Abstract: Arrhythmia can be detected by carefully studying the electrocardiogram (ECG) and the distortions in the QRS complex. Since the appearance of the distorted beats, the indicators of arrhythmia, may occur randomly with respect to time and span over a large time interval, an automated classification mechanism may reduce the tedium in identifying and isolating these beats. This paper proposes an arrhythmia classifier based on probabilistic neural networks. The data is derived from MIT-BIH arrhythmia database. The classifier is designed to classify ten different types of beats, where the difference is based on morphology of the beat. Ten statistical morphological parameters are computed from the training dataset and they form the feature vector for the PNN training. The proposed classifier performs quite well with an average classification accuracy of 98.1%, average sensitivity of 0.9810, average specificity of 0.9978, average positive prediction rate as 0.981, average false prediction rate of 0.002 and average classification rate of 0.9962. The main advantage of using PNN is that it requires no training and a new class category can be added without major modifications to the network.

VMLHST: Development of an Efficient Novel Virtual Reality ML Framework with Haptic Feedbacks for Improving Sports Training Scenarios

Abstract: This paper presents the development of a novel virtual reality (VR) machine learning (ML) framework that incorporates haptic feedback to improve sports training scenarios. The framework uses You Look Only Once (YoLo) for object detection, and combines it with ensemble learning to analyze the performance of athletes in a simulated environment and provide real-time feedbacks. The system includes haptic feedback devices that are controlled via Grey Wolf Optimization (GWO) to simulate the physical sensation of a real-world sports scenario, allowing athletes to experience the sensation of force, impact, and movements. The proposed system was tested using a group of professional athletes who participated in various sports, including football, basketball, and tennis. The athletes were asked to perform various training scenarios in the virtual environment, and their performance was compared with their real-world performance levels. The results showed that the proposed system improved the athletes' performance significantly, as they were able to receive immediate and accurate feedback on their movements, and the haptic feedback provided a realistic sensory experience that enhanced their trainings. The proposed research has the potential to revolutionize sports training by providing athletes with an efficient and effective way to improve their performance in a set of safe and controlled environments. The system can be customized to suit various sports and training scenarios, and the ML algorithms can be trained on large datasets to improve their accuracy and effectiveness. The incorporation of haptic feedback provides a unique and realistic experience, making the training more engaging and effective under real-time scenarios. The proposed system showcased an accuracy 93.5%, with 3.5% higher precision, and 4.9% higher recall than existing models, which has the potential to enhance athletic performance and revolutionize the way athletes train for different sports.

An efficient fpga overlay for color transformation function using high level synthesis

Abstract: Image Processing is a significantly desirable in commercial, industrial, and medical applications. Processor based architectures are inappropriate for real time applications as Image processing algorithms are quite intensive in terms of computations. To reduce latency and limitation in performance due to limited amount of memory and fixed clock frequency for synthesis in processor-based architecture, FPGA can be used in smart devices for implementing real time image processing applications. To increase speed of real time image processing custom overlays (Hardware Library of programmable logic circuit) can be designed to run on FPGA fabric. The IP core generated by the HLS (High Level Synthesis) can be implemented on a reconfigurable platform which allows effective utilization of channel bandwidth and storage. In this paper we have presented FPGA overlay design for color transformation function using Xilinx’s python productivity board PYNQ-Z2 to get benefit in performance over a traditional processor. Performance comparison of custom overlay on FPGA and Processor based platform shows FPGA execution yields minimum computation time.

Survey of Stochastic Computing for Hardware implementation

Abstract: Stochastic Computing (SC) is another substitute way to deal with digital information. With ongoing advances in Internet of Thigs (IOT) and wearable gadgets it is interesting to implement Artificial Neural Network (ANN), Image Processing applications on reconfigurable platform. However, several elements and very complicated connections are usually required, which leads to more hardware and power consumption. Stochastic Computing (SC) meet stringent requirement in embedded and mobile IOT devices. SC is an alternate approach to deterministic computing that treats binary data as probabilities. SC employs very low-complexity arithmetic units. SC is popular in immensely parallel systems and is exceptionally tolerant to soft errors. SC based hardware execution achieves low hardware foot print, low power and energy competent circuits while preserving high accuracy. This paper surveys SC's the low cost, low complexity, error tolerant, alternative to binary computing

JPEG2000-Based Semantic Image Compression using CNN

Abstract: Some of the computer vision applications such as understanding, recognition as well as image processing are some areas where AI techniques like convolutional neural network (CNN) have attained great success. AI techniques are not very frequently used in applications like image compression which are a part of low-level vision applications. Intensifying the visual quality of the lossy video/image compression has been a huge obstacle for a very long time. Image processing tasks and image recognition can be addressed with the application of deep learning CNNs as a result of the availability of large training datasets and the recent advances in computing power. This paper consists of a CNN-based novel compression framework comprising of Compact CNN (ComCNN) and Reconstruction CNN (RecCNN) where they are trained concurrently and ideally consolidated into a compression framework, along with MS-ROI (Multi Structure-Region of Interest) mapping which highlights the semiotically notable portions of the image. The framework attains a mean PSNR value of 32.9dB, achieving a gain of 3.52dB and attains mean SSIM value of 0.9262, achieving a gain of 0.0723dB over the other methods when compared using the 6 main test images. Experimental results in the proposed study validate that the architecture substantially surpasses image compression frameworks, that utilized deblocking or denoising post- processing techniques, classified utilizing Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index Measures (SSIM) with a mean PSNR, SSIM and Compression Ratio of 38.45, 0.9602 and 1.75x respectively for the 50 test images, thus obtaining state-of-art performance for Quality Factor (QF)=5.

Computer-aided Arrhythmia Diagnosis with Bio-signal Processing: A Survey of Trends and Techniques

Abstract: Signals obtained from a patient, i.e., bio-signals, are utilized to analyze the health of patient. One such bio-signal of paramount importance is the electrocardiogram (ECG), which represents the functioning of the heart. Any abnormal behavior in the ECG signal is an indicative measure of a malfunctioning of the heart, termed an arrhythmia condition. Due to the involved complexities such as lack of human expertise and high probability to misdiagnose, long-term monitoring based on computer-aided diagnosis (CADiag) is preferred. There exist various CADiag techniques for arrhythmia diagnosis with their own benefits and limitations. In this work, we classify the arrhythmia detection approaches that make use of CADiag based on the utilized technique. A vast number of techniques useful for arrhythmia detection, their performances, the involved complexities, and comparison among different variants of same technique and across different techniques are discussed. The comparison of different techniques in terms of their performance for arrhythmia detection and its suitability for hardware implementation toward body-wearable devices is discussed in this work.

A Novel Feature Level Fusion for Heart Rate Variability Classification Using Correntropy and Cauchy-Schwarz Divergence.

Abstract: Heart rate variability (HRV) analysis has become a widely used tool for monitoring pathological and psychological states in medical applications. In a typical classification problem, information fusion is a process whereby the effective combination of the data can achieve a more accurate system. The purpose of this article was to provide an accurate algorithm for classifying HRV signals in various psychological states. Therefore, a novel feature level fusion approach was proposed. First, using the theory of information, two similarity indicators of the signal were extracted, including correntropy and Cauchy-Schwarz divergence. Applying probabilistic neural network (PNN) and k-nearest neighbor (kNN), the performance of each index in the classification of meditators and non-meditators HRV signals was appraised. Then, three fusion rules, including division, product, and weighted sum rules were used to combine the information of both similarity measures. For the first time, we propose an algorithm to define the weights of each feature based on the statistical p-values. The performance of HRV classification using combined features was compared with the non-combined features. Totally, the accuracy of 100% was obtained for discriminating all states. The results showed the strong ability and proficiency of division and weighted sum rules in the improvement of the classifier accuracies.

Evaluation of Novel Entropy-Based Complex Wavelet Sub-bands Measures of PPG in an Emotion Recognition System

Abstract: Nowadays, the increasing demand for human–computer interface applications shows the social need to provide an accurate, intelligent emotion recognition system. Computer-aided emotion recognition using physiological signals remains a challenging task in “affective computing”. In this paper, the entropy-based complex wavelet sub-bands (ECWS) measures are suggested for the classification of 14 emotion categories from the photoplethysmograph (PPG). PPG data available at DEAP (a Database for Emotion Analysis using Physiological signals) were selected when subjects were watching the fun, exciting, joy, happy, cheerful, love, sentimental, melancholy, sad, depressing, mellow, hate, shock, and terrible music videos. Using the dual-tree complex wavelet transforms, each PPG signal was decomposed into six levels. Four entropy measures, including approximate entropy, sample entropy, permutation entropy, and the improved multi-scale permutation entropy, were extracted from each sub-band coefficient. Then, the normalized ECWS features were input to the probabilistic neural network. The role of sigma adjustment was also considered in classifier performance. The results indicated the accuracy rates of 92 to 100% for the classification of 14 emotional states. The maximum accuracy rate was 100% for sigma < 0.25. Our findings establish the proposed system as a superior framework compared to the state-of-the-art PPG emotion recognition tool.

ADDHard: Arrhythmia Detection with Digital Hardware by Learning ECG Signal

Abstract: Anomaly detection in Electrocardiogram (ECG) signals facilitates the diagnosis of cardiovascular diseases i.e., arrhythmias. Existing methods, although fairly accurate, demand a large number of computational resources. Based on the pre-processing of ECG signal, we present a low-complex digital hardware implementation (ADDHard) for arrhythmia detection. ADDHard has the advantages of low-power consumption and a small foot print. ADDHard is suitable especially for resource constrained systems such as body wearable devices. Its implementation was tested with the MIT-BIH arrhythmia database and achieved an accuracy of 97.28% with a specificity of 98.25% on average.

Diagnosing Atrial Septal Defect from Electrocardiogram with Deep Learning

Abstract: The heart murmur associated with atrial septal defects is often faint and can thus only be detected by chance. Although electrocardiogram examination can prompt diagnoses, identification of specific findings remains a major challenge. We demonstrate improved diagnostic accuracy realized by incorporating a proposed deep learning model, comprising a convolutional neural network (CNN) and long short-term memory (LSTM), with electrocardiograms. This retrospective observational study included 1192 electrocardiograms of 728 participants from January 1, 2000, to December 31, 2017, at Tokyo Women's Medical University Hospital. Using echocardiography, we confirmed the status of healthy subjects—no structural heart disease—and the diagnosis of atrial septal defects in patients. We used a deep learning model comprising a CNN and LTSMs. All pediatric cardiologists (n = 12) were blinded to patient groupings when analyzing them by electrocardiogram. Using electrocardiograms, the model’s diagnostic ability was compared with that of pediatric cardiologists. We assessed 1192 electrocardiograms (828 normally structured hearts and 364 atrial septal defects) pertaining to 792 participants. The deep learning model results revealed that the accuracy, sensitivity, specificity, positive predictive value, and F1 score were 0.89, 0.76, 0.96, 0.88, and 0.81, respectively. The pediatric cardiologists (n = 12) achieved means of accuracy, sensitivity, specificity, positive predictive value, and F1 score of 0.58 ± 0.06, 0.53 ± 0.04, 0.67 ± 0.10, 0.69 ± 0.18, and 0.58 ± 0.06, respectively. The proposed method is a superior alternative to accurately diagnose atrial septal defects.