Towards Next-Generation Cardiac Care in Resource-Constrained and High-Risk Scenarios

Chandra, B S and Jana, Soumya (2019) Towards Next-Generation Cardiac Care in Resource-Constrained and High-Risk Scenarios. PhD thesis, Indian institute of technology Hyderabad.

Full text not available from this repository. (Request a copy)

Abstract

Universal healthcare is a cherished objective of the human society. Unfortunately, various obstacles to achieve this objective presently appear insurmountable. To begin with, functioning of healthcare facilities traditionally requires interaction among medical professionals, medical equipment and medical information management. However, such facilities often number far fewer than necessary, especially in developing countries, mainly due to dearth of medical professionals and high equipment cost among other reasons. Consequently, medical diagnosis, treatment and treatment response monitoring are inaccessible and/or unaffordable to vast segments of population belonging to remote and economically disadvantaged communities. In this dissertation, we take a step towards realizing a low-cost method of delivering reliable healthcare to remote communities under severe resource constraints. Another significant healthcare challenge arises in managing high-risk patients. To improve their chance of survival, these individuals require continuous assessment of their health. However, it is generally not feasible to have medical professionals to monitor such patients continuously, and so one usually depends on automated health monitoring and alert generation tools. In this context, a home-based continuous monitoring service appears attractive, provided impending health conditions can be detected fast and accurately. In this dissertation, we proposed a technique that lays the technological groundwork for the desired service. An even higher risk is involved in managing patients, who have recently suffered (or been assessed to have strong likelihood of suffering) a catastrophic health event, and are continuously being monitored at intensive care units (ICUs). There, the objective is to reliably detect the onset of an adverse medical event based on a number of vital parameters. To improve upon the state of the art, we propose a sophisticated method that significantly reduces the false alarm rate in ICUs without compromising the rate of adverse event detection. In this dissertation, we investigate technological solutions to problems related to resource-constrained healthcare and critical care scenarios in the context of cardiovascular diseases (CVDs), which account for more than 30% of the global mortality. In managing CVDs, the electrocardiogram (ECG) signal indicating the electrical activity of the heart plays an indispensable role. A conventional telecardiology system that simply records and transmits user ECG signals to a professional diagnostic facility appears to be an attractive in remote communities. However, with severe constraints on infrastructure in such communities, operation of a conventional telecardiology system could be infeasible. Against this backdrop, we proposed a novel two-tier telecardiology framework, where constraints on resources, such as power and bandwidth, are met by compressively sampling ECG signals, and transmitting only the anomalous signals. Specifically, we designed practical compressive classifiers based on inherent properties of ECG signals, such as self-similarity and periodicity. Using such classifiers, we realize a resource-constrained telecardiology system that operates under severe power and bandwidth constraints. Next, to manage certain CVD patients requiring continuous monitoring, we proposed an inexpensive yet accurate home-based ECG monitoring service. Specifically, we seek to provide point-of-care monitoring of premature ventricular contractions (PVCs), high frequency of which could indicate the onset of potentially fatal arrhythmia. To this end, we developed a dictionary-based algorithm that reduces not only the overall bandwidth requirement, but also the physician’s workload by localizing anomalous beats. Subsequently, we consider high-risk patient care, and propose convolutional neural network (CNN) based methods for detecting various critical arrhythmias. Finally, to manage highest-risk group patients that are admitted to ICUs, multiple physiological signals usually including ECG, photoplethysmogram (PPG) and arterial blood pressure (ABP), are continuously monitored to detect onset of adverse health events. Yet, an ICU alarm is traditionally generated by treating each signal independently, and raised even in innocuous situations. Although rarely missing a catastrophic event, the traditional scheme significantly increases the frequency of false alarms resulting in wasted resources and decreased staff responsiveness to alarms. We propose algorithms based on CNNs that can be trained to identify desired arrhythmic conditions from aforementioned physiological signals. In the proposed approach, generalizable CNN modules underlie various processing units including heartbeat detectors and specific arrhythmia detectors and validators. We demonstrate the efficacy of our algorithms on standard databases as well as hidden datasets. Also, while the present work proves the soundness of the proposed concepts based on relevant databases, one still requires to build end-to-end systems, and evaluate those on extensive field data collected from patients. In fact, we envision a next-generation healthcare system that would assist medical professionals in reducing their time and effort for managing patients, be agile in initiating life-saving interventions, provide critical services at affordable cost even under resource constraints, and have improved reliability in critical care scenarios. Although we focused on cardiac care and associated vital signals, we anticipate our approach to extend to broader healthcare issues dealing with additional physiological signals. Indeed, our sparsity and learning based techniques potentially apply to an even broader class of natural signals, such as scene analysis and computer vision.

[error in script]

IITH Creators: