Source-linked AI summary
Fog Data: Enhancing Telehealth Big Data Through Fog Computing
Harishchandra Dubey, Jing Yang, Nick Constant, Amir Mohammad Amiri, Qing Yang, Kunal Makodiya
TL;DR
Telehealth wearable sensors produce large, varied datasets that are challenging to process quickly. Fog Data analyzes sensor data at an onsite Fog gateway; in the speech case, DTW achieved more than 99% data reduction while performing close to CLIP.
Problem
Telehealth collects large volumes of varied sensor data that require real-time, fast processing, posing challenges for healthcare big-data handling.
Source: §1 · p. 1Method
Results
In the speech case, DTW achieved more than 99% data reduction while performing close to CLIP.
Source: §6 · p. 5Takeaways & Limitations
The authors conclude Fog Data has potential to reduce logistics requirements for telehealth applications.
Source: §7 · p. 6Higher-compression bzip2 and 7-zip programs require more CPU power and RAM, making them unsuitable for embedded Fog processors.
Source: §4.3 · p. 4
Abstract
from arXiv · showhide
The size of multi-modal, heterogeneous data collected through various sensors is growing exponentially. It demands intelligent data reduction, data mining and analytics at edge devices. Data compression can reduce the network bandwidth and transmission power consumed by edge devices. This paper proposes, validates and evaluates Fog Data, a service-oriented architecture for Fog computing. The center piece of the proposed architecture is a low power embedded computer that carries out data mining and data analytics on raw data collected from various wearable sensors used for telehealth applications. The embedded computer collects the sensed data as time series, analyzes it, and finds similar patterns present. Patterns are stored, and unique patterns are transmited. Also, the embedded computer extracts clinically relevant information that is sent to the cloud. A working prototype of the proposed architecture was built and used to carry out case studies on telehealth big data applications. Specifically, our case studies used the data from the sensors worn by patients with either speech motor disorders or cardiovascular problems. We implemented and evaluated both generic and application specific data mining techniques to show orders of magnitude data reduction and hence transmission power savings. Quantitative evaluations were conducted for comparing various data mining techniques and standard data compression techniques. The obtained results showed substantial improvement in system efficiency using the Fog Data architecture.
1. INTRODUCTION
Fog Data addresses telehealth big-data and battery-efficiency challenges by analyzing sensor time series on an embedded Fog computer before sending necessary data to the cloud. A working prototype and experiments demonstrated data reduction in telehealth applications.
1. INTRODUCTION
- Telehealth must process large, heterogeneous, multimodal, distributed medical datasets in real time while keeping battery-powered body sensor networks running; storage and transmission consume significant energy.Wearable sensors enable non-invasive, unobtrusive, 24/7 health monitoring, while preliminary analytics can reduce the data that must be stored and transmitted.Source: §1 · p. 1
- The service-oriented Fog Data architecture analyzes wearable and ambient sensor time series onsite, filtering data before an energy-efficient embedded Fog computer sends necessary information to the cloud.It is designed to support in-home collection, storage, and analysis of heterogeneous, multimodal, distributed, and scalable data for person-centered health monitoring.Source: §1 · p. 2
- Experiments with an Intel Edison prototype and real patient and open-source datasets showed an order of magnitude in data reduction, with case studies in speech disorders and ECG monitoring.The paper reports that the prototype was connected to wearable telehealth systems.Source: §1 · p. 2
2. RELATED WORKS
Telehealth extends health monitoring into non-clinical settings, but sensor-rich data and noisy raw uploads make cloud-based analysis costly and inefficient. Fog computing is presented as a nearby gateway for handling data and supporting telehealth applications.
2.1. TeleHealth and Medical Big Data
- Wearable telehealth generates complex medical data, including errors and unwanted signals, making raw-data uploads to the cloud inefficient, time-consuming, and expensive.The section describes telehealth monitoring in homes and other non-clinical settings and notes that complexity increases for multimodal data.Source: §2.1 · p. 2
2.2 Fog Computing as a Smart Gateway
- A Fog gateway connects to wearable sensors, processes incoming data, maintains queryable clinical information, and can protect data and user identity.The paper presents its Fog infrastructure and two case studies as demonstrations of Fog computing for wearable telehealth data.Source: §2.2 · p. 3
3. THE ARCHITECTURE OF FOG DATA
Fog Data links wearable sensors, an onsite Fog gateway, and a cloud server to collect, process, and store timestamped health data. The gateway extracts clinically relevant features or patterns for application-dependent transfer, while the cloud handles storage and more complex analysis.
3. THE ARCHITECTURE OF FOG DATA
- Fog Data cascades from a body sensor network through an onsite Fog gateway to cloud storage, with the gateway accumulating data and extracting features or mining patterns on demand.Depending on the healthcare application, raw data, extracted features, or time-series pattern indexes can be sent to the cloud. Because Fog resources are limited, its algorithms must be computationally simple yet clinically relevant; complex diagnosis and monitoring analysis runs in the cloud, and transferred data is deleted from the Fog computer.Source: §3 · p. 3
3.1 Information Flow in Fog Data Architecture
- Fog Data stores wearable-sensor measurements as timestamped time series and processes raw data into timestamped features or patterns.Associated .xls logs record acquisition times and factors such as remaining battery power, supporting access by physicians, patients, and engineers and helping debug software.Source: §3.1 · p. 3
- Wearable sensors can send data through nearby tablets or smartphones, avoiding custom hardware; the authors used their validated EchoWear framework to acquire speech data for case study 1.The architecture supports modalities including speech, ECG, and other clinical vital signs, using devices such as smartwatches and smart clothing-based ECG sensors.Source: §3.1 · p. 3
4. METHODS AND ALGORITHMS
The section describes methods implemented on Intel® Edison for healthcare-data analysis, including time-series pattern mining, speech processing, compression, local data reduction, and power measurement.
4.1 Dynamic Time Warping
- DTW matches time-series patterns despite differing lengths and phase, using an adjacency matrix and shortest-path search.The passage contrasts DTW with Euclidean distance, which fails to detect similarity between similar but out-of-phase series.Source: §4.1 · p. 4
4.2 Clinical Speech Processing Chain (CLIP)
- CLIP estimates loudness from speech root mean squared power and pitch with low-pass-filtered AMDF, followed by median filtering.The amplitude-based loudness estimate and computationally inexpensive AMDF method were chosen for the Fog processor’s limited resources; median filtering refines AMDF estimates that can double or halve the true pitch.Source: §4.2 · p. 4
4.3 Compression
- Fog Data uses GNU zip on the Fog processor and decompresses data in the cloud; bzip2 and 7-zip offer higher compression ratios but require more CPU and RAM –.The resource demands of bzip2 and 7-zip make them unsuitable for embedded Fog processors.Source: §4.3 · p. 4
4.4 Data and Bandwidth Reduction
- Local processing on the Fog processor reduces healthcare-data transmission and storage, with DTW, clinical speech processing, and Pan-Tompkins ECG processing implemented on Intel® Edison.DTW was programmed in C, while CLIP used a Python front end with a C back end; the methods can be used on processors similar to Intel® Edison.Source: §4.4 · p. 4
4.5 Power Consumption
- 862 mW for DTW computations and 937 mW for encrypted data transmission through Wi-Fi were measured on Intel® Edison.DTW runs occasionally and briefly; the authors report that a standard 9V battery can supply the system for at least 4 hours. They also state that Fog Data consumes less power than Raspberry Pi.Source: §4.5 · p. 4
5. EXPERIMENTAL SETUP
Experiments used an Intel Edison as a low-power Fog platform running C/Python programs, including a software DTW implementation. DTW used clinically selected reference patterns and a waiting window that transmitted only the most similar detected pattern.
5. EXPERIMENTAL SETUP
- The experiments used an Intel Edison Fog platform with a 500 MHz dual-core Atom CPU, 1 GB memory, WiFi, and 35 mW idle consumption with WiFi open.The board ran ubilinux and Python 2.7.2; low-level C programs were packaged with Python wrappers.Source: §5 · p. 4
- DTW was implemented in C using UCR Suite and libdtw, with the UCR Suite implementation selected for real-time time-series processing.The authors describe UCR Suite as one of the fastest software implementations; accelerated alternatives required special hardware support.Source: §5 · p. 4
- Clinically selected DTW reference patterns and a waiting window limited transmission to the most similar detected pattern; the speech case used a 40 ms reference segment.When multiple similarity patterns were detected within the window, only the most similar one was sent to the back-end server.Source: §5 · p. 5
6. FOG DATA: CASE STUDIES
The case studies apply Fog Data to smartwatch speech monitoring and ECG monitoring, processing sensor data on an Intel® Edison Fog computer and sending extracted information to the cloud.
6.1 Case Study 1: Telehealth of Speech Motor Disorders
- For speech, CLIP’s loudness-and-fundamental-frequency features achieved the highest data reduction; DTW exceeded 99%, while GNU zip was reversible but reduced less.CLIP and DTW are lossy and cannot reconstruct the original speech files; GNU zip can reconstruct them in the cloud.Source: §6.1 · p. 5
- The speech case sends smartwatch signals to an Intel® Edison Fog computer, which extracts loudness and pitch features and pattern indices for cloud access by SLPs customizing exercises.The system supports remote speech treatment through EchoWear.Source: §6.1 · p. 5
- On Intel® Edison, 2160 ECG samples took 1 second to process, and QRS detection stayed below 15 seconds across 16.24–36.45 kB datasets.Execution time increased with file size.Source: §6.1 · pp. 5–6
- For ECG monitoring, Intel® Edison processes MIT-BIH data using Pan-Tompkins QRS detection, then sends DTW pattern indices or GNU zip-compressed time series to the cloud.The MIT-BIH Arrhythmia data is cited as.Source: §6.1 · p. 5, §6 · p. 5
- DTW-based pattern mining reduced ECG data by more than 98% in most cases, while GNU zip compression reduced it by around 91%.The difference in reduction efficiency was smaller than for speech data.Source: §6.1 · p. 5, Figure 5 · p. 6
7. CONCLUSIONS
Fog Data was validated for speech disorders and ECG, with potential to reduce telehealth logistics, cloud storage, and edge transmission power. Future work includes additional speech features and in-home EchoWear validation.
7. CONCLUSIONS
- Fog Data was validated for speech disorders and ECG, with potential to reduce telehealth logistics, cloud storage, and edge-device transmission power.The paper identifies reduced logistics requirements, cloud storage, and edge transmission power as potential benefits.Source: §7 · p. 6
- Future speech-disorder analysis could add shimmer, jitter, and sensory pleasantness to loudness and fundamental frequency.These additional features can help quantify perceptual speech quality.Source: §7 · p. 6
- Future work could use Fog Data for in-home validation studies of EchoWear, developed in.Source: §7 · p. 6