Você está na página 1de 9

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 13, NO.

3, MAY 2009

351

A Telehealth Architecture for Networked Embedded Systems: A Case Study in In Vivo Health Monitoring
Foad Dabiri, Member, IEEE, Tammara Massey, Member, IEEE, Hyduke Noshadi, Hagop Hagopian, C. K. Lin, Robert Tan, Jacob Schmidt, and Majid Sarrafzadeh, Fellow, IEEE

AbstractThe improvement in processor performance through continuous breakthroughs in transistor technology has resulted in the proliferation of lightweight embedded systems. Advances in wireless technology and embedded systems have enabled remote healthcare and telemedicine. While medical examinations could previously extract only localized symptoms through snapshots, now continuous monitoring can discretely analyze how a patients lifestyle affects his/her physiological conditions and if additional symptoms occur under various stimuli. We demonstrate how medical applications in particular benet from a hierarchical networking scheme that will improve the quantity and quality of ubiquitous data collection. Our Telehealth networking infrastructure provides exibility in terms of functionality and the type of applications that it supports. We specically present a case study that demonstrates the effectiveness of our networked embedded infrastructure in an in vivo pressure application. Experimental results of the in vivo system demonstrate how it can wirelessly transmit pressure readings measuring from 0 to 1.5 lbf/in2 with an accuracy of 0.02 lbf/in2 . The challenges in biocompatible packaging, transducer drift, power management, and in vivo signal transmission are also discussed. This research brings researchers a step closer to continuous, real-time systemic monitoring that will allow one to analyze the dynamic human physiology. Index TermsCollaborative technologies in health care delivery, information science and technology, teleconsultation, telemedicine, telesurgery.

I. INTRODUCTION

ECENT advances in the electronics industry and wireless communication have enabled innovative domains of applications to evolve. Embedded processors and systems have become widely used in peoples everyday life in various applications ranging from mobile communication to automotive industries to medical applications. There has been several studies and system developments for real-time health monitoring through wireless technologies. Some examples of application-specic medical systems are

Manuscript received February 9, 2008; revised May 1, 2008 and July 22, 2008. First published January 23, 2009; current version published May 6, 2009. This work was supported in part by Microsoft Research. This paper was presented in part at the 4th International Workshop on Body Sensor Networks in Aachen, Germany, in 2007. F. Dabiri, T. Massey, H. Noshadi, H. Hagopian, and M. Sarrafzadeh are with the Department of Computer Science, University of California, Los Angeles, CA 90095 USA (e-mail: dabiri@cs.ucla.edu; tmassey@cs.ucla.edu; hyduke@cs.ucla.edu; shaitani@gmail.com; majid@cs.ucla.edu). C. Lin, R. Tan, and J. Schmidt are with the Department of Biomedical Engineering, University of California, Los Angeles, CA 90095 USA (e-mail: schmidt@seas.ucla.edu. Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TITB.2009.2013248

wristbands for measuring pulse, body temperature, galvanic skin reactions, and electromyography (EMG) data [1], [2]; chest and arm belts for physiologic monitoring [3], [4]; shoes for gait analysis [5], [6]; and photo plethysmographic ring sensors [7]. The development of standards in practice, technology, and information processing are crucial to the success of wireless Telehealth applications. Well-designed standards will ensure interoperability, integrity, and compliance among various medical devices. Assuredly, the lack of standards will lead to devices that are prohibitively expensive, complex, unreliable in operation, and present a large educational and training burden for healthcare providers. We have used our infrastructure in various applications. As an illustrative instance, we present a minimally invasive implantable pressure sensing system that actively monitors longterm physiological changes in real time. Specically, we investigate pressure changes in the upper urinary tract per degree of obstruction. Extensive work has been done in passive pressure monitoring where an incoming radio signal induces the pressure measurement. The earliest pressure capsules were developed by Jacobson and Machay [8] and Farrar et al. [9] in 1957 using a passive telemetering capsule that used the motion of the iron near the coil to determine internal pressure and temperature. Later, an implantable passive eye transistor measured pressure [10]. Fonseca et al. [11] investigated passive pressure sensors that did not contain a power supply or an active circuitry for high-temperature environments. Passive measurements enable the construction of extremely miniature devices, but have a low range and must be close to an emitting radio signal in order for measurements to be collected. Recently, Fonseca et al. [12] developed a passive pressure sensor for acute uses with liquid crystal polymers and chronic uses using ceramic fabrication. The previous work on in vivo systems (such as [13]) introduce technologies for in vivo monitoring; our contribution is to utilize these technologies and integrate it with features such as recongurability and customization to make a general purpose ubiquitous health monitoring with minimum expert supervision. Compared with the aforementioned research, our Telehealth infrastructure and in vivo active pressure monitoring system has the following unique features: 1) the continuous, active monitoring of pressure within the upper urinary track to determine how ones lifestyle affects physical symptoms; 2) the aggregation of data to an online database that stores information for later analysis of the symptoms;

1089-7771/$25.00 2009 IEEE


Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

352

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 13, NO. 3, MAY 2009

Fig. 1. System components of the active pressure system: a PDA serving as an OBT, a catheter collecting sensor data, a wireless transmitter, and a lightweight microprocessor or Mednode.

3) the remote in vivo reconguration of the software; 4) dynamic power and energy management. II. HEALTH MONITORING INFRASTRUCTURE In this section, we will go over individual parts of our proposed network structure for real-time health monitoring. In the proposed system, each patient has lightweight wireless modules that form on-body networks and utilize local on-body servers that communicate with other on-body sensors or the medical enterprises. Fig. 2 in Section III illustrated this general hierarchical architecture. The hierarchical architecture contains sensors, Mednodes, on-body terminals (OBTs), and central server. Fig. 1 illustrates the components of our architecture implemented in an in vivo pressure monitoring platform. Our hierarchical architecture is similar to other architectures proposed [14][16], but our architecture has reconguration capabilities and unique power saving techniques for lightweight embedded medical systems. Mednodes are the main components of our proposed wearable architecture. A Mednodes is a stand-alone component that consists of a processing unit, a sensor board, and a local power supply. We have used Mica2Dots and Mica2s from CrossBow [17] as the main processing unit in Mednodes. These lightweight embedded systems are called motes in general and are equipped with Atmel ATmega 128L microprocessor with multichannel analog-to-digital converters (ADCs). They also use 868/ 916 MHz multichannel radio transceiver for wireless communication. Various sensors are interfaced to these channels through a sensor board. The sensor board operates as an interface to both excite a passive sensor and feed the sensed signal to an ADC channel. In some scenarios, a signal conditioning circuit is embedded onto the sensor board as well. Mednodes are capable of wireless communication and support two-wired communication protocols: interintegrated circuit (I2C) and universal asynchronous receiver/transmitter (UART). A coin cell battery is used to supply the Mednodes with power that immediately raises the power consumption challenges in system design. We describe these challenges and propose methodologies to overcome these challenges shortly. The Mednodes communicate with an OBT. The OBT is a personal server that is a programmable lightweight system, such as a cell phone or a PDA. These devices collect data from sensors, store, or process data, and transmit data to an enterprise. We use different off-the-shelf devices as part of our system, such

as PocketPCs, cell phones, and portable multimedia devices (e.g., iPod). In particular, we tested our system on two different PocketPCs from HP (iPaq) that had WiFi connectivity and connected to a local area network. Also, they support global system for mobile communication (GSM) and can be used as a cell phone. In our system, the OBT acts as an intermediate connector between patient and physician. The communication protocol used by Mednodes is 802.15.4 (Zigbee compliance). On the other hand, the OBT, which is a handheld device, does not support 802.15.4. Therefore, a gateway radio connects to the OBT and collects data from Mednodes wirelessly, and transmits it to an OBT via a serial connection (UART). The OBT opportunistically transfers data to a remote server by either using WiFi or Bluetooth. In more recent versions of Mednodes, we equipped them with RN-24 Bluetooth modules [18] to enable direct communication between Mednodes and on-body network (OBN). The nal leg in this design hierarchy is the central server. The server collects the data received from multiple OBTs and functions as a central storage unit. Moreover, extensive data analysis and processing are scheduled to be performed at the central server. One of the unique features of the server in our prototype is its capability to monitor the functionality of individual patients and recongure OBTs and associated Mednodes according to real-time needs. More details on reconguration techniques are described in [19]. Arguably, much of the emphasis in past work on healthcare applications of wearable sensors and devices has been driven by the hardware design and communication infrastructure, and less attention has been paid to the problems of integrating such information into the medical enterprise. Thus, ideally, the server would execute several tasks including: 1) data incorporation into a patients electronic medical record (EMR); 2) additional processing and/or signal analysis given the additional computational processing power available on servers; and 3) the reconguration of resources on the handheld device and Mednodes using interactions of healthcare providers and the EMR itself (i.e., in a feedback loop). III. HIERARCHICAL NETWORK STRUCTURE The structure of the entire network can be described by breaking down the network into two layers. The lower of the two layers are the branches in the network that represent the OBN and the wireless embedded sensor network that is physically located on the body. The remaining layer of the network is the network of OBNs or personal area networks (PANs). Fig. 2 illustrates a general overview of the networking hierarchy. It starts from individual Mednodes that form the BANs, and then leads to the PANs, and nally, the central server/medical enterprise. The PAN allows both the communication between independent OBNs and the communication of individual OBNs to the server. The server module contains the server gateway, which all OBNs must establish a connection to communicate with the server (Fig. 2). The network has also the capability of connecting independent OBNs. These OBNs have a exible interface that allows various sensors from various manufacturers to connect up to the node. For example, the

Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

DABIRI et al.: TELEHEALTH ARCHITECTURE FOR NETWORKED EMBEDDED SYSTEMS: A CASE STUDY IN In Vivo HEALTH MONITORING

353

Fig. 2.

High-level diagram of the proposed network.

interface for sensors has a connector that can be removed which can handle sensors that require a different number of channels. Nodei represents the ith sensor node (or Mednode) that is part of the OBN. The OBT is the gateway node between any Nodei and the upper layers of the network. The OBT connects to other OBNs and to the server gateway. The hierarchical network structure proposed here is a reliable, efcient, and recongurable design that promotes stability and network expansion. By using an OBT as a gateway between the upper layers of the network and the lower (OBN) layer, unnecessary complication and power consumption are minimized greatly. In this highly organized hierarchical design, any node can connect to any OBN from the outside and any node is able to send data to the outside. In this infrastructure, there is no need for an amalgamated network that would use a signicantly larger number of resources and would dwarf attempts of network expansion. We will elaborate more on OBNs in the next section. A. Expansion of the Network As mentioned before, our hierarchical design allows expansion in at least two dimensions. The rst dimension (from the lower layer) is the expansion of the quantity of sensor nodes by any integer k. If n represents the number of sensor nodes in any given OBN, then the system should allow for a total of n + k nodes. Our system begins with n = 0, and then, adds one node at a time until any number of nodes n + k are added. The only limitation presented by our communication protocol provides the upper bound n + k < 255, where 255 is the largest nonreserved integer representable as an 8-bit number. Note that this limitation is dynamic; a simple change in protocol will allow for a larger number of nodes to be added. The communication protocols does not rely on the upper limit on n; in other words, increasing this limit is just a matter of parameter update and packet exchanging protocols remain intact. In the OBN model, the rst dimension of expansion of nodes follows the next protocol. Nodex would like to join an OBN where three nodes are already connected. The OBT must rst recognize Nodex s request to join the OBN. The OBT then forward the request to the server. The server has the knowledge of the previous joined nodes, Nodes 1 through 3. The server assigns Nodex , an identication number to distinguish any future reconguration commands sent to the node and sensor data received from the node. The OBT will then relate the new ID number to Nodex and congure it to identify itself using its newly assigned

ID. This process can be repeated to add additional nodes to the OBN. The second dimension of expansion is the addition of a new OBN module to the network. To add a new OBN, there exists only one new protocol. A nonhierarchical network would have increased the number of protocols required to add a new branch to the network. However, in our system, the addition of the OBN protocol is inherently parallel to the earlier OBT communication protocol. In this layer of the architecture, a regular sensor node must be differentiated from an OBT. As a new OBT requires a new ID number to distinguish it from other nodes on an OBN, a new OBN also requires unique identier. When a new OBN joins the network, it sends a request to the OBN to be added to the network. The OBN forward this request to the server. The server, having kept track of the previous OBNs added, assigns a unique OBN ID number to the OBN. Once the server accepts the request and returns a new ID, the OBN congures itself to use that new ID number to distinguish itself from the other OBNs. IV. NETWORK COMMUNICATION PROTOCOLS This section discusses the various command protocols currently implemented by our network and the theory behind the minimum number of unique command protocols required to have a fully functional patient sensor network. Currently, our communication protocol consists of three basic forms of communication. Two communication protocols are for unique identication requests, and the third protocol is for sensor data transfer. This is the absolute minimum number of independent command structures necessary to implement the network in our design. A. OBN Formation Protocols In order to appropriately identify each layer of the network (i.e., sensor nodes and OBN), we must rst distinguish a request coming from a sensor node and a request coming from an OBN in order to assign ID numbers unique to the appropriate branch. In order to congure a node, we specify two things: a branch (i.e., OBN) and a sensor node within that branch. In order to accommodate these two dimensions, there are two different types of requests in this network. The lower level aspects of a packet and protocol are described shortly. Each type of packet corresponds to three different basic requests: incoming data from sensor node to server, an OBT ID request, or a sensor node ID request. For each packet, there is a slashed eld at the beginning. This slash represents a reserved integer used as a frame marker that marks the beginning of the packet. The type 1 packet is the format of incoming data from a sensor node. A type 1 header contains the packet type, patient ID (unique identier for individual OBNs), mote ID (unique identier for sensor nodes), data type, and sequence number. A type 2 packet contains a request for a new patient ID. Type 2 packets prevents relevant information from being passed from a new OBT. A type 3 packet is a request from a new sensor node asking to be assigned the next available ID in its OBN. All earlier three packet types are

Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

354

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 13, NO. 3, MAY 2009

Fig. 4.

Message type for OBTl protocols.

to its nal destination. As an alternate method to form ad hoc networks, our system can utilize Bluetooth to connect itself to other Bluetooth enabled authorized nodes in the network. It is desired for applications to have the exibility to choose their wireless medium. Once an OBT connects with another OBT, the type of connection is determined based on variety of factors, such as intention, power, policy, and availability.
Fig. 3. Ad hoc networks created by individual OBTs where each OBT represents an OBN.

C. Communication Protocols for the OBT Each OBN is composed of two autonomous components: Mednodes and OBTs. Each Mednode sends a wireless embedded system that is connected to several sensors through ADC ports, and they have the ability to collect sensor data and share them with other nodes in the network wirelessly. Each Mednode is sending two type of packets: advertise message (AdvMsg) and data packet. Initially, each Mednode must obtain a unique Mednode ID and patient ID from the OBT. Once a Mednode is added to a body network, it broadcasts an AdvMsg that passes the nodes unique network address to the OBT and introduces the sensor mote as a new Mednode on the body. To maintain reliability in the network, there is a time-out variable associated with each AdvMsg sent. The time out occurs when a Mednode does not receive a response from the OBT during the time-out interval when the AdvMsgs are sent. The ability to set the patient ID and Mednode ID dynamically allows the network to be exible so that at any time a new sensor can be added to an OBN. Also, Mednodes can be swapped between different patients by restarting them in the new environment. The Mednode starts transmitting data packets as soon as it receives a proper (nonezero) Mednode ID and patient ID. The types of messages are illustrated in Fig. 4. OBTs are composed of two different components: a wireless sensor platform (Mica2 or Mica2dot) and a PDA. Currently, our OBTs work with two different interfaces: 1) PDA communication (Fig. 5) and 2) PC Gateway communication (Fig. 6). If the OBT is in the PC Gateways wireless range, the terminal would communicate with the PC Gateway; otherwise, it would communicate with the PDA, as previously discussed. The mote on the OBT is responsible for maintaining a patient ID for the OBN, maintaining and assigning Mednode IDs for Mednodes in the OBN, and redirecting the packets from Mednodes to the PC or PDA. Same as Mednode, every OBT should request a patient ID upon start-up. Therefore, an AdvMsg would be sent to the PC Gateway at the request of a patient node. Moreover, the OBT requests a new patient ID from the PC Gateway upon the arrival of an AdvMsg from the new Mednode.

sent from the network to the server, but two of them require a response. There are two responses that are sent from the OBT to the sensor node. Type 2 responses are from the OBT requesting a unique patient ID. Type 3 responses are from a sensor node requesting a unique mote ID to the OBN. Due to the openended design in our communication protocol, we can readily expand our library of commands to almost any conguration. For example, the server can specify an OBN, a sensor node, and a new command of packet type x to change the sampling rate. This level of congurability allows real-time changes to be made to the system. B. Communication Between OBTs The OBTs carried by patients have the capability of communicating by other similar devices in the environment. The OBTs have the capability of forming an ad hoc network with each other and transfer data when policy dictates (see Fig. 3). Each mobile OBT can join the ad hoc network by requesting an Internet protocol (IP) address specic to the ad hoc network. The request will be made to a dynamic host conguration protocol (DHCP) server, which can be centralized on the local server or distributed on the existing mobile devices. Once the request is received by DHCP server, a new IP address will be given to that particular node. The OBTs send requests to members of the network. This message can be the permission to disseminate data in case of emergency. Once permission is granted, the client node will initiate a connection with the target device and start sending messages. User datagram protocol (UDP) is a connectionless protocol. Therefore, there will be no guarantee that the message has been transferred successfully. To overcome this uncertainty, the receiver OBTs will send acknowledgment message to the sender with each received message containing the received packet number. The packets of data will eventually get delivered

Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

DABIRI et al.: TELEHEALTH ARCHITECTURE FOR NETWORKED EMBEDDED SYSTEMS: A CASE STUDY IN In Vivo HEALTH MONITORING

355

D. Server Architecture The server module described in previous sections is actually composed of a software server and a gateway mote, referred to earlier as the Server Gateway (SG). In our system, we used a Mica2 Mote, which is similar to the sensor nodes and the OBTs. The gateway mote allows multiple patients to wirelessly communicate with the server at the same time. This gateway mote will eventually connect to a medical enterprise that may be using previously designed protocols, such as HL7 or ISO/IEEE 11073. Data from our infrastructure can easily be extracted and put into the HL7 data structure, a standard for the exchange of electronic medical data. Standards, such as ISO/IEEE 11073 for medical devices, can also t within our infrastructure. Interoperability between various standards is possible due to the fact that our infrastructure can easily extract the data or encapsulate the data with headers for the format to be consistent with another standard.

E. Wireless Peer-to-Peer Networking: Where do We Stand? In this section, we briey elaborate on the differences between our approach and the state-of-the-art peer-to-peer (P2P) networking methodologies. P2P) networking is among very well studies elds of networking and communication. There has been several recent researches exploring different architectures for P2P wireless networking for various application domains. [20][22]. Where all these results investigate novel methodologies for coordination and communication in wireless networks, the involvement of body area networks in our proposed infrastructure poses new challenges. Since by nature of our target systems and applications, hierarchy becomes an inevitable part of the network topology, adapting existing methods lacks efciency, as described in this section. In hierarchical approaches such as [23], communicating units share almost the same functionalities that are then clustered in a hierarchical manner. One of the most important features of our health monitoring infrastructure is the fact that on-body communication devices (i.e., MedNodes), OBTs, and the base station are very different components that introduce coordination challenges in a heterogenous wireless networking. Our proposed architecture treats each layer of hierarchy based on local functionalities. Furthermore, it automatically emulates the concept of plug-andplay in P2P networking, which is hardly addressed by related researches.

Fig. 5. Message exchanging over time for the scenario where the PDA is the destination.

Fig. 6. Message exchanging over time for the scenario where the gateway is the destination.

V. CASE STUDY: DEVELOPMENT OF AN In Vivo ACTIVE PRESSURE MONITORING Localized obstructions between the kidney and the urethra cause elevated pressures in both the upper urinary track (kidney, renal pelvis, and ureter) and lower urinary track (bladder, prostate, and urethra). Overlooked acute and chronic elevated pressure in the urinary track often leads to increased risk of infection, the formation of kidney stones, and irreversible damage to the bladder and the kidneys (if untreated for an extensive period of time) [24].

Then, it sends back its own patient ID and the new Mednode ID to the new Mednode in a packet called ConformMsg (see Fig. 4) The OBT is also responsible for propagating the data packets arriving from the Mednodes to the PC Gateway.

Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

356

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 13, NO. 3, MAY 2009

Kidney stones, ureter strictures, tumors, and uretopelvic junction obstruction block the upper urinary track. A symptom of urinary blockage is elevated pressure. In the lower tract, elevated pressures often results in involuntary bladder spasms and a lack of bladder control. Lack of bladder control and involuntary spasms are often sources of embarrassment and lifestyle impediment for the millions of men and women with urinary incontinence [24]. An urodynamics examination of the lower urinary tract is the customary diagnostic test for urinary track blockage. This examination is performed by placing a transurethral catheter into the bladder and the rectum. While supine, the patients bladder is lled with saline over the course of 2 h. This articial stimulus causes considerable discomfort to the patient and often results in an inaccurate snapshot of pressure in the urinary track. If one could accurately determine the level of pressure within the urinary track, surgical and pharmaceutical measures can reduce the pressure and avoid harmful complications that result from prolonged elevated pressure [25], [26]. To address the problem of localized pressure monitoring, an implantable active pressure sensor has been developed for the continuous measurement of elevated pelvic and urethral renal pressures. This ambulatory sensor evaluates patients with urethral obstruction and has applications in additional pressure applications, such as intracranial. The overall goal of the in vivo pressure monitoring system is to actively gather and distribute information on the pressure within the upper urinary track in a manner that is extremely fault tolerant. The technical challenges in developing the system are the design and fabrication of the transducer, the design and construction of the catheter, and the development of a recongurable program that can remotely update the implanted sensor node with future code modications. Portions of this case study with initial results has been published in [27]. Unlike passive sensors, our active sensor has its own internal power supply that generates an outgoing signal. Active pressure monitoring has the benets of increased transmission range and ubiquity in deployment. Our system can more easily communicate with current devices already being carried, such as PDAs. The drawbacks of battery lifetime and size are justiable in applications where active communication of medical conditions are necessary. Additionally, the data are transmitted to a PDA that uploads the data to an online database. This allows for the later analysis of the data to determine patterns in how ones lifestyle affects the internal urinary track pressure and the longterm progression of urinary track blockage.

Fig. 7.

Pigtail in vivo catheter.

A. Pressure Transducers The conversion of pressure into an electrical signal is achieved by the physical deformation of strain gauges that are bounded to the diaphragm of the pressure transducer and wired into a Wheatstone bridge conguration. Pressure applied to the pressure transducer produces a deection of the diaphragm that introduces strain to the gauges. The strain will produce an electrical resistance change proportional to the pressure. A miniature original equipment manufacturer (OEM) pressure die measuring 0.65 mm 0.65 mm senses the pressure. The desired range of pressure sensing in our application was 0 to 1 lbf/in2 with accuracy of 0.02 lbf/in2 . Resistance change corresponds to this range and results in a very small peak-to-peak voltage difference, 6 mV. Since the pressure range in our application is 0 to 1 lbf/in2 , the voltage span is approximately 6 mV. The small degree of voltage can be considerably affected by environmental noise. Even the uctuation of the power supply voltage can cause considerable error in the circuit if it is not designed carefully. The precise signal conditioning circuit is responsible for the following. 1) Stabilizing excitation voltage: The sensor excitation voltage needs to be steadied at 3 V. 2) Electrical noise reduction: Considerable noise is generated within the circuit and must be accurately ltered out. 3) Voltage amplication: The voltage range must be amplied to fully utilize the 0 to 3 voltage range. 4) Offset voltage removal: The sensors bridge is not symmetric when no pressure is applied. This lack of symmetry results in an offset voltage that reduces the voltage swing between 0 to 1.5 lbf/in2 . The differential amplier buffers the sensor from the rest of the circuit and partially removes the offset voltage. The second amplifying stage magnies the voltage by a factor of approximately 53. The highfrequency noise is signicantly reduced with the low-pass lter. A voltage regulator is responsible to drive both the sensor and signal conditioning circuit with a stable voltage. The circuit detects a 1/50 lbf/in2 pressure change.

VI. ACTIVE PRESSURE MONITORING SYSTEM The system design for continuous in vivo monitoring of intraluminal pressures in the urinary tract includes integrated novel sensors, wireless miniaturized sensor nodes, and proven implant packaging technologies pioneered by Minimed Medtronic (see Fig. 7). The research focus is to safely implant these devices for 48 h of usage in animals and subsequently humans. B. Catheter Design The diameter, materials, and stiffness of the catheter have been carefully designed for usability, biocompatibility, and maneuverability for in vivo operation (Fig. 8). This is a seven French catheter (equivalent to 2.33 mm) constructed from four platinumiridium conductors wound in a helix around a high-tensile polyester core. The conductors are individually

Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

DABIRI et al.: TELEHEALTH ARCHITECTURE FOR NETWORKED EMBEDDED SYSTEMS: A CASE STUDY IN In Vivo HEALTH MONITORING

357

processor. In vivo code updates are essential in creating a system that can be recongured after deployment. E. Biocompatible Packaging Unsatisfactory packaging degrades the performance of the sensor and can result in device failure or a severe immunogenic response from the subject. Most researchers package their implantable sensors with only silicone dipping or polyethylene. However, most silicones are not designed to be used internally and polyethylene has been shown to attract immunogenic cells. With the knowledge of the pitfalls of conventional packaging schemes, we specically designed the package to be a modular platform for catheter-based implantable sensors. We packaged our system with two dual passivation layers: polyethylene and medical grade encapsulant. To combat the volatile environments in the body and the harsh chemicals in postprocessing, we evaporated a thin layer of polyethylene to protect the sensors. With its low surface energy, polyethylene deposition is not limited by line-of-sight, and therefore conforms to any geometry on the microscale. Polyethylene forms a pinhole-free conformal coating of a thickness as small as 0.03 m. Polyethylene also has high resistance to permeation, and solvent absorption and strengthens wire bonds at interfaces. Medical grade silicone, on the other hand, has excellent bulk property and malleability to surface properties. Silicone also has a long history of biological and biomedical applications. The pliability, surface property exibility, and biocompatibility make silicone particularly attractive for long-term medical device packaging material. F. Experimental Results We tested the system in dry-laboratory setting and also implanted it inside a pig. The pig implantation was done using normal medical procedures that were approved by UCLA Institutional Review Board (IRB). We rst tested the pressure sensor to verify its accuracy and determine the repeatability of the system. We put the pressure sensor in a pressure chamber and modied the pressure between 0 and 1.5 lbf/in2 . We also covered the sensor node and the data also showed a linear t. Voltagepressure dependency is modeled as a linear function with less than 0.004 in standard deviation. To combat drift, the signal was changed to a square wave form from a dc voltage. Then the duty cycle was minimized to a short time period that allowed the pressure to be measured accurately. The drift in each sensor is characterized by a trend line and the absolute values from various parameters are compared. Our system addresses drift by lowering the duty cycle of the excitation voltage input. There is a 10% drift within 800+ at a frequency of 50 k and a duty cycle of 20% (Table I). During the animal study, two calibrated sensors were implanted in a porcine model: one was placed inside the peritoneal cavity while the other was implanted within the bladder with the former providing a reference. In both cases, the lead tip was inserted into the aforementioned space and the main body of the device was kept in a subcutaneous pocket. The sensor in the peritoneal cavity had its battery drain out after 13 h of

Fig. 8.

Catheter design cross section.

polytetrauoroethylene (PTFE) insulated and the catheter body is sheathed in silicone rubber. A detailed picture of the catheter is shown in the upper righthand corner of Fig. 1. Rubber tube of barium-lled material improves radiopacity. A molded and strain-relieved coaxial connector (tip and ring) is afxed at one end of the catheter. The conductors are soldered to bonding pads on a printed circuit board (PCB) substrate and encapsulated with biocompatible epoxy. In addition, to relieve the anticipated forces exerted on the system during packaging and implantation, the monitoring system includes stress relief measures to ensure additional robustness (Fig. 1) C. Power and Communication The two main consumers of power in this system are: 1) sensor bridge and signal conditioning circuit and 2) wireless communication. The power source in the implantable platform is a li-polymer battery that cannot be recharged or changed after it has been implanted. Therefore, a simple power management strategy was adopted for the voltage regulator. The voltage regulator drives the sensor and the signal conditioning circuit through control signals from the CPU. The system excites the sensor for 20 ms during each sampling period (1 Hz frequency). After the sampling period, the device goes in inactive mode and power dissipation is greatly reduced when the circuit is completely shut off. This simple pulse excitation method reduces the power consumption signicantly. The multichannel radio has a data rate of 38.5 kbps, and consumes 89 mW when transmitting data and 33 mW when receiving data. Sampled data are stored locally on the Mica2Dot and transmitted to the PDA every 30 s to reduce power consumption and header overhead. The pulse excitation method and accumulated data transmissions increase the lifetime of the system from 6 h to 48 h. D. Remote In Vivo Software Reconguration After the sensor is implanted, the software may need to be modied. Therefore, an operating system designed specically for embedded systems called sensor operating system (SOS) was used [28]. This message passing system severs the ties between the core operating system and individual applications or modules. A module measuring the pressure was created that could be loaded or removed at run-time without interrupting the core operating system. A lightweight medical processor can transmit a program or module to be executed on the in vivo

Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

358

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 13, NO. 3, MAY 2009

TABLE I ANALYSIS OF DRIFT TOLERANCE WITH VARIOUS FREQUENCIES AND DUTY CYCLES

Fig. 10. Detrusor pressure from the experimental results showing a muscle contraction. The detrusor pressure data are not affected by abdominal straining, gas, and abdominal contents, and depicts the activities in the bladder more accurately.

Fig. 9. Typical in vivo pressure data from a porcine model. The in vivo pressure for the bladder and peritoneal cavity are shown where the sharp peak indicates a muscle contraction in the pig.

operation, and the sensor in the bladder continued to operate for 48 h after implantation, at which point the pig was sacriced and both devices were explanted. An autopsy determined there was no adverse reaction to the implanted sensors. Toward the end of the experiment, there were difculties with high noise due to a small amount of leakage near the connector between the sensor and embedded device. However, clinically relevant data were obtained for 2 h. Fig. 9 depicts the bladder voiding prole. As urine is collected, the compliant nature of the surrounding tissue permits the bladder to expand without signicant pressure change. However, as the bladder gets full, muscle contractions force the urine into the ureter, which is seen as a rapid rise in pressure and a corresponding sharp drop. Detrusor pressure is calculated by subtracting the measured peritoneal cavity pressure from the bladder pressure 10. Abdominal straining, gas, and abdominal contents do not affect the detrusor pressure, and therefore, detrusor pressure more accurately represents the actual activities that occur within the bladder (Fig. 10). G. Discussion We demonstrated the development of a minimally invasive, implantable system capable of continuous in vitro and in vivo data collection. This implantable telemetry system can collect comprehensive lifestyle data from a patient for an extended period of time, and therefore, is a useful research and clinical management tool. Several obstacles presented themselves during our investigation. During our characterization process, we discovered that upon voltage excitation, our raw sensor output has a slight drift. Though the drift within 48 h was minimal, during a longer implanted period, the drift may reduce the usefulness of

the system in the clinic. To meet the 48 h specication, we used a high capacity, miniature prole lithium polymer cell and used power conservation techniques. To further extend the lifetime of the system, we are investigating mechanisms to recharge the cell, to further reduce the duty cycle, and to optimize software excitation or circuit components. Furthermore, we will also expand the range of sensor functions to measure pulse, saturated oxygen, and temperature. This will be done through the in-house fabrication of these sensors on the same substrate as the pressure sensor. This sensor platform will eventually enable the collection of massive physiological data in multiple parts of the body. VII. FINAL REMARKS AND FUTURE DIRECTIONS In this paper, we described a hierarchical network architecture that enables real-time health monitoring. This network infrastructure includes a lightweight sensing/processing wireless modules placed on the body as the lower level, and the medical enterprise servers as a higher level. The network communication protocol presented will allow for improved diagnosis, medical treatment, and monitoring in Telehealth applications. One of the main features of our architecture is the dynamic nature in the sense that OBNs can easily be tailored for individuals and dynamically updated with new specications. As an illustrative example, we implemented the architecture in a minimally invasive, implantable system capable of continuous in vivo data collection. Direct, continuous, and minimally invasive pressure measurements shed light on the diagnosis of conditions resulting from elevated pressure in the urinary track. Our in vivo pressure monitoring system enables the creation of therapeutic guidelines for the lower urinary track. Moreover, the future coupling with an actuator system would enable timely delivery of local therapy, and revolutionize the treatment of conditions and collection of comprehensive lifestyle data from a patient for an extended period of time. We must emphasize that our experiments have been limited to only two trials because of the surgical procedure. We are in the process of conducting the third experiment. This is one of the limitation of our case study, but we believe that the current experimental results have proven the concept.

Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

DABIRI et al.: TELEHEALTH ARCHITECTURE FOR NETWORKED EMBEDDED SYSTEMS: A CASE STUDY IN In Vivo HEALTH MONITORING

359

The proposed general purpose hierarchical architecture for health monitoring can enable ubiquitous and personalized wearable and implantable monitoring. We believe that many medical monitoring applications can t into this hierarchical architecture. Currently, we are developing this infrastructure and are including different applications. In future, we plan to perform clinical tests and study this in a large group of patients to quantify the effectiveness of this architecture. REFERENCES
[1] K. Ouchi, T. Suzuki, and M. Doi, Lifeminder: A wearable healthcare support system using users context, in Proc. ICDCSW, 2002, pp. 791 792. [2] O. Amft, G. Troster, P. Lukowicz, and C. Schuster, Sensing muscle activities with body-worn sensors, in Proc. BSN 2006, Proc. Int. Workshop Wearable Implantable Body Sens. Netw. (BSN 2006), IEEE Computer Society, Washington, DC, pp. 138141. [3] G. Troster, P. Lukowicz, U Anliker, S. J. Schwartz, and R.W. DeVaul, The weararm modular, low-power computing core, IEEE Micro, vol. 21, no. 3, pp. 1628, May/Jun. 2001. [4] E. Jovanov, A. ODonnell Lords, D. Raskovic, P. G. Cox, R. Adhami, and F. Andrasik, Stress monitoring using a distributed wireless intelligent sensor system, IEEE Eng. Med. Biol. Mag., vol. 22, no. 3, pp. 4955, May/Jun. 2003. [5] S. J. Morris and J. A. Paradiso, Shoe-integrated sensor system for wireless gait analysis and real-time feedback, in Proc. 2nd Joint Eng. Med. Biol. 24th Annu. Conf. Annu. Fall Meeting Biomed. Eng. Soc. EMBS/BMES, 2002, pp. 24682469. [6] M. R. Popovic, V. Dietz, M. Morari, I. P. I. Pappas, T. Keller, and S. Mangold, A reliable gyroscope-based gait-phase detection sensor embedded in a shoe insole, IEEE Sensors J., vol. 4, no. 2, pp. 268274, Apr. 2004. [7] R. A. Sokwoo Rhee, H. H. Asada, P. Shaltis, and R. C. Hutchinson, Mobile monitoring with wearable photoplethysmographic biosensors, IEEE Eng. Med. Biol. Mag., vol. 22, no. 3, pp. 2840, May/Jun. 2003. [8] B. Jacobson and R. Mackay, Endoradiosonde, Proc. IEEE, vol. 179, no. 4572, pp. 12391240, Jun. 1957. [9] J. T. Farrar, V. K. Zworykin, and J Baum, Pressure-sensitive telemetering capsule for study of gastrointestinal motility, Science, vol. 126, no. 3280, pp. 975976, 1957. [10] C. C. Collins, Miniature passive pressure transensor for implanting in the eye, IEEE Trans. Biomed. Eng., vol. BME-14, no. 2, pp. 7483, Apr. 1967. [11] J. White, M. Fonseca, J. Kroh, and M. Allen, Flexible wireless passive pressure sensors for biomedical applications, presented at the Tech. Dig. Solid-State Sensor, Actuator, Microsyst. Workshop, Hilton Head Island, SC, 2006. [12] M. von Arx, M. Fonseca, J. English, and M. Allen, High temperature characterization of ceramic pressure sensors, in Proc. Transducers, 2001, pp. 486489.

[13] A. Arena, C Caccamo, P Valdastri, A Menciassi, and P Dario, An implantable telemetry platform system for in vivo monitoring of physiological parameters, IEEE Trans. Inf. Technol. Biomed., vol. 8, no. 3, pp. 271278, Sep. 2004. [14] C. Otto, E. Jovanov, A. Milenkovic, and P. C. de Groen, A wireless body area network of intelligent motion sensors for computer assisted physical rehabilitation, J. NeuroEng. Rehabil., vol. 2, no. 6, 2005. [15] D. Koutsouris, S. Dembeyiotis, and G. Konnis, Integrating legacy medical data sensors in a wireless network infrastructure, presented at the 27th Annu. Int. Conf. Eng. Med. Biol. Soc., Shanghai, China, Sep. 2005. [16] M. Welsh, R. R. Juang, A. Alm, T. Gao, and D. Greenspan, Vital signs monitoring and patient tracking over a wireless network, in Proc. 27th Annu. Int. Conf. Eng. Med. Biol. Soc., 2005, pp. 102105. [17] (2007/2008). [Online]. Available: http://www.xbow.com/ [18] (2007/2008). [Online]. Available: http://www.rovingnetworks.com/ modules.htm [19] H. Hagopian, T Massey, F Dabiri, H Noshadi, and M. Sarrafzadeh, Lightweight medical bodynets, presented at the 2nd Int. Conf. Body Area Netw., Florence, Italy, 2007. [20] J. Hill, M. Horton, R. Kling, and L. Krishnamurthy, The platforms enabling wireless sensor networks, Commun. ACM, vol. 47, no. 6, pp. 41 46, 2004. [21] W. Chen and S. Cai, Ad hoc peer-to-peer network architecture for vehicle safety communications, IEEE Commun. Mag., vol. 43, no. 4, pp. 100 107, Apr. 2005. [22] R Schollmeier, I Gruber, and F Niethammer, Protocol for peer-to-peer networking in mobile environments, in Proc. 12th Int. Conf. Comp. Commun. Netw. (ICCCN 2003), pp. 121127. [23] S Banerjee and S Khuller, A clustering scheme for hierarchical control in multi-hop wireless networks, in Proc. IEEE 12th Annu. Joint Conf. IEEE Comput. Commun. Soc. (IFOCOM 2001), pp. 10281037. [24] G. Lose, Urethral pressure measurementProblems and clinical value, Scand. J. Urol. Nephrol., vol. 35, no. 6, pp. 6166, 2 Mar. 2001. [25] M Aga and Y Esashi, Biomedical microsystems for minimally invasive diagnosis and treatment, Proc. IEEE, vol. 92, no. 1, pp. 98114, Jan. 2004. [26] B. Rose. (2007/2008). Diagnosis of urinary tract obstruction and hydronephrosis [Online]. Available: www.uptodate.com [27] F. Dabiri, T. Massey, R. Tan, M. Sarrafzadeh, M. Srivastava, P. G. Schulam, J.Schmidt C. K Lin, D Jea, and C. D Montemagno, The development of an in-vivo active pressure monitoring system, presented at the 4th Int. Workshop Body Sens. Netw. (BSN 2007), Aachen, Germany. [28] C.-C. Han, R. Kumar, R. Shea, E. Kohler, and M. Srivastava, A dynamic operating system for sensor nodes, in Proc. MobiSys 2005, Proc. 3rd Int. Conf. Mobile Syst., Appl., Serv., ACM, New York, 2005, pp. 163176.

Authors photographs and biographies not available at the time of publication.

Authorized licensed use limited to: Ramabhadran Sampath. Downloaded on October 11, 2009 at 10:26 from IEEE Xplore. Restrictions apply.

Você também pode gostar