Solarillion Foundation, Chennai, India
Abstract:Blood pressure (BP) is one of the most influential bio-markers for cardiovascular diseases and stroke; therefore, it needs to be regularly monitored to diagnose and prevent any advent of medical complications. Current cuffless approaches to continuous BP monitoring, though non-invasive and unobtrusive, involve explicit feature engineering surrounding fingertip Photoplethysmogram (PPG) signals. To circumvent this, we present an end-to-end deep learning solution, BP-Net, that uses PPG waveform to estimate Systolic BP (SBP), Mean Average Pressure (MAP), and Diastolic BP (DBP) through intermediate continuous Arterial BP (ABP) waveform. Under the terms of the British Hypertension Society (BHS) standard, BP-Net achieves Grade A for DBP and MAP estimation and Grade B for SBP estimation. BP-Net also satisfies Advancement of Medical Instrumentation (AAMI) criteria for DBP and MAP estimation and achieves Mean Absolute Error (MAE) of 5.16 mmHg and 2.89 mmHg for SBP and DBP, respectively. Further, we establish the ubiquitous potential of our approach by deploying BP-Net on a Raspberry Pi 4 device and achieve 4.25 ms inference time for our model to translate the PPG waveform to ABP waveform.
Abstract:Atrial fibrillation (AF) is the most prevalent cardiac arrhythmia worldwide, with 2% of the population affected. It is associated with an increased risk of strokes, heart failure and other heart-related complications. Monitoring at-risk individuals and detecting asymptomatic AF could result in considerable public health benefits, as individuals with asymptomatic AF could take preventive measures with lifestyle changes. With increasing affordability to wearables, personalized health care is becoming more accessible. These personalized healthcare solutions require accurate classification of bio-signals while being computationally inexpensive. By making inferences on-device, we avoid issues inherent to cloud-based systems such as latency and network connection dependency. We propose an efficient pipeline for real-time Atrial Fibrillation Detection with high accuracy that can be deployed in ultra-edge devices. The feature engineering employed in this research catered to optimizing the resource-efficient classifier used in the proposed pipeline, which was able to outperform the best performing standard ML model by $10^5\times$ in terms of memory footprint with a mere trade-off of 2% classification accuracy. We also obtain higher accuracy of approximately 6% while consuming 403$\times$ lesser memory and being 5.2$\times$ faster compared to the previous state-of-the-art (SoA) embedded implementation.
Abstract:Fake news is fabricated information that is presented as genuine, with intention to deceive the reader. Recently, the magnitude of people relying on social media for news consumption has increased significantly. Owing to this rapid increase, the adverse effects of misinformation affect a wider audience. On account of the increased vulnerability of people to such deceptive fake news, a reliable technique to detect misinformation at its early stages is imperative. Hence, the authors propose a novel graph-based framework SOcial graph with Multi-head attention and Publisher information and news Statistics Network (SOMPS-Net) comprising of two components - Social Interaction Graph (SIG) and Publisher and News Statistics (PNS). The posited model is experimented on the HealthStory dataset and generalizes across diverse medical topics including Cancer, Alzheimer's, Obstetrics, and Nutrition. SOMPS-Net significantly outperformed other state-of-the-art graph-based models experimented on HealthStory by 17.1%. Further, experiments on early detection demonstrated that SOMPS-Net predicted fake news articles with 79% certainty within just 8 hours of its broadcast. Thus the contributions of this work lay down the foundation for capturing fake health news across multiple medical topics at its early stages.
Abstract:Smartphones contain information that is more sensitive and personal than those found on computers and laptops. With an increase in the versatility of smartphone functionality, more data has become vulnerable and exposed to attackers. Successful mobile malware attacks could steal a user's location, photos, or even banking information. Due to a lack of post-attack strategies firms also risk going out of business due to data theft. Thus, there is a need besides just detecting malware intrusion in smartphones but to also identify the data that has been stolen to assess, aid in recovery and prevent future attacks. In this paper, we propose an accessible, non-intrusive machine learning solution to not only detect malware intrusion but also identify the type of data stolen for any app under supervision. We do this with Android usage data obtained by utilising publicly available data collection framework- SherLock. We test the performance of our architecture for multiple users on real-world data collected using the same framework. Our architecture exhibits less than 9% inaccuracy in detecting malware and can classify with 83% certainty on the type of data that is being stolen.
Abstract:Continuous monitoring of cardiac activity is paramount to understanding the functioning of the heart in addition to identifying precursors to conditions such as Atrial Fibrillation. Through continuous cardiac monitoring, early indications of any potential disorder can be detected before the actual event, allowing timely preventive measures to be taken. Electrocardiography (ECG) is an established standard for monitoring the function of the heart for clinical and non-clinical applications, but its electrode-based implementation makes it cumbersome, especially for uninterrupted monitoring. Hence we propose SeismoNet, a Deep Convolutional Neural Network which aims to provide an end-to-end solution to robustly observe heart activity from Seismocardiogram (SCG) signals. These SCG signals are motion-based and can be acquired in an easy, user-friendly fashion. Furthermore, the use of deep learning enables the detection of R-peaks directly from SCG signals in spite of their noise-ridden morphology and obviates the need for extracting hand-crafted features. SeismoNet was modelled on the publicly available CEBS dataset and achieved a high overall Sensitivity and Positive Predictive Value of 0.98 and 0.98 respectively.
Abstract:Freezing of Gait (FoG) is a common gait deficit among patients diagnosed with Parkinson's Disease (PD). In order to help these patients recover from FoG episodes, Rhythmic Auditory Stimulation (RAS) is needed. The authors propose a ubiquitous embedded system that detects FOG events with a Machine Learning (ML) subsystem from accelerometer signals . By making inferences on-device, we avoid issues prevalent in cloud-based systems such as latency and network connection dependency. The resource-efficient classifier used, reduces the model size requirements by approximately 400 times compared to the best performing standard ML systems, with a trade-off of a mere 1.3% in best classification accuracy. The aforementioned trade-off facilitates deployability in a wide range of embedded devices including microcontroller based systems. The research also explores the optimization procedure to deploy the model on an ATMega2560 microcontroller with a minimum system latency of 44.5 ms. The smallest model size of the proposed resource efficient ML model was 1.4 KB with an average recall score of 93.58%.
Abstract:Human computer interaction facilitates intelligent communication between humans and computers, in which gesture recognition plays a prominent role. This paper proposes a machine learning system to identify dynamic gestures using tri-axial acceleration data acquired from two public datasets. These datasets, uWave and Sony, were acquired using accelerometers embedded in Wii remotes and smartwatches, respectively. A dynamic gesture signed by the user is characterized by a generic set of features extracted across time and frequency domains. The system was analyzed from an end-user perspective and was modelled to operate in three modes. The modes of operation determine the subsets of data to be used for training and testing the system. From an initial set of seven classifiers, three were chosen to evaluate each dataset across all modes rendering the system towards mode-neutrality and dataset-independence. The proposed system is able to classify gestures performed at varying speeds with minimum preprocessing, making it computationally efficient. Moreover, this system was found to run on a low-cost embedded platform - Raspberry Pi Zero (USD 5), making it economically viable.