Network Centric Therapy is in essence a manifestation of the Internet of Things for the domain of healthcare and biomedical engineering. Central to this concept is the utility of wearable and wireless systems comprised of inertial sensors, such as the accelerometer and gyroscope. Wearable and wireless inertial sensor systems can objectively quantify human movement characteristics. Smartphones and portable media devices in particular are equipped with inertial sensors to enable quantified recording of their accelerometer and gyroscope signals. Furthermore, locally wireless inertial sensor nodes can convey inertial sensor data streams to devices, such as smartphones, portable media devices, and tablets, for pending wireless transmission to Cloud derived data storage systems for post-processing. This configuration enables optimized therapy that is network centric to occur with the patient and therapist situated remotely, and even potentially on opposite sides of the world. Two broad themes for Network Centric Therapy are addressed: gait and associated reflex quantification and quantification of movement disorder status and assessment of deep brain stimulation efficacy.
Keywords: Smartphone, portable media device, accelerometer, gyroscope, wearable device, wireless, wearable and wireless system, machine learning, reflex response, gait, movement disorder, Parkinson’s disease, Essential tremor, deep brain stimulation, Network Centric Therapy
Network Centric Therapy refers to a concept developed by LeMoyne and Mastroianni during the publication of the book “Wearable and Wireless Systems for Healthcare I: Gait and Reflex Response Quantification”. In essence, this perspective represents a transformative inflection regarding therapy, rehabilitation, and intervention techniques, for which the logistical and interactive constraints between a patient and clinician are liberated through the presence of the Internet. The implications are that the patient, clinician, and even the associated post-processing resources can literally be geographically segmented anywhere in the world [1, 2, 3]. For example, a patient in remote Africa could receive undergo rehabilitation therapy from experts in the United States of America with machine learning post-processing to augment diagnostic acuity from Switzerland.
Network Centric Therapy constitutes a third transitional phase of the continuous evolutionary pathway for rehabilitation and prescribed medical intervention techniques [1, 2, 3]. The first phase characterized by LeMoyne and Mastroianni represents the application of ordinal scale techniques . These ordinal scale techniques involve the observation of a highly skilled clinician. Upon observing the response of the patient to prescribed scenario, the clinician applies an expert, although subjective, perspective based on the interpretation of ordinal scale criteria [4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15].
The transition to the second phase characterized by LeMoyne and Mastroianni involves the application of instrumented devices for providing a quantified interpretation of the patient’s status . These systems are generally reserved and limited to a clinical laboratory [5, 6, 16, 17]. This presents a dilemma regarding the permeability of logistics between clinical team and patient, for which the allocation of limited medical resources may present and issue. The preliminary developments of the third transitional phase, even in the nascent stages, appear to substantially alleviate these concerns.
The origins of the third transitional phase known as Network Centric Therapy derive from the successful research, development, testing, and evaluation of human movement characteristics through smartphones and associated portable media devices [1, 2, 3, 18]. Preliminary, experimentation was conducted by LeMoyne et al. during 2010 for the quantification of gait and Parkinson’s disease tremor using the smartphone as a wearable and wireless accelerometer platform [19, 20, 21, 22]. A most notable observation was the experiment was conducted in metropolitan Pittsburgh, Pennsylvania with the trial data conveyed wirelessly as an email attachment using the Internet to post-processing resources in metropolitan Los Angeles, California, which are trans-continentally situated in the United States of America. The implication was the experimental and post-processing could be situated anywhere in the world [1, 18, 23, 24, 25, 26].
With the capabilities of the smartphone and associated portable media device constituting a wearable and wireless inertial sensor platform, there are many applications for the biomedical community for objectively quantifying patient status. Smartphones and portable media devices functioning as wearable and wireless systems have been extensively applied to the domains of gait and reflex quantification [1, 2, 18, 23, 24, 25, 26]. Regarding the domain of movement disorders, such as Parkinson’s disease and Essential tremor, smartphones have been applied to ascertain the efficacy of advanced intervention techniques, such as deep brain stimulation [25, 26, 27, 28, 29, 30]. With the objectively quantified inertial signal data, advanced post-processing techniques, such as machine learning have been successfully applied [31, 32]. These recent achievements and future perspectives of the prominence of Network Centric Therapy representing the Internet of Things for the healthcare and the biomedical community are examined.
Preliminary first phase involving ordinal scales for quantification of gait, reflex response, and movement disorder
Regarding gait, the ordinal scale technique applies clinically a quantified perspective regarding rehabilitation status [4, 11, 12]. Multiple ordinal scales strategies are available at the discretion of the clinician for evaluating the characteristics of a patient’s gait status . However, an apparent issue is that progressive modifications in gait can be inherently minute. The interpretation of status is subjective in nature, for which the experience of the clinician can influence the evaluation process .
Tendon reflexes, such as with respect to the patella, are highly associated with gait [6, 34]. Regarding reflex quantification, two prevalent ordinal approaches for quantifying tendon reflex response are the NINDS five-point scale and Mayo Clinic nine-point scale [8, 9]. However, the reliability of these scales is a topic of contention [6, 9, 10, 34].
Movement disorder ordinal scale methodologies are contextual to the type of movement disorder being diagnosed, such as Parkinson’s disease and Essential tremor [14, 15]. An issue regarding this technique is that with the multiple ordinal scale strategies, there does not exist an established means of translating the ordinal scales in an interchangeable manner . Furthermore, the reliable application of ordinal scales methodologies is proportional to the expertise of the clinician observing the movement disorder .
An additional observation is the patient must schedule an appointment with the clinician for an evaluation. Especially regarding neurological disorders that are temporally variable, the effective snapshot of a clinical appointment may not optimally characterize the patient’s true status. Also, a patient must travel to the clinical setting, which introduces a logistic consideration into the process. These issues are considerably relieved with the introduction of wearable and wireless inertial sensor systems with access to the Internet.
Second phase applying instrumented quantification of gait, reflex response, and movement disorder
Regarding gait analysis and quantification there are six prevalent conventional instrumented devices:
- Optical motion camera
- Force plate
- Metabolic analysis
- Foot switch
The utility of these devices is the ability to objectively evaluate gait [37, 38]. These devices can provide a basis for evaluating therapy response, for which quantified feedback can provide extremely valuable acuity for the efficacy of a comprehensive rehabilitation strategy [37, 39]. Tandem integration of gait analysis equipment, such as the kinematic data acquired by a optical motion camera and kinetic data derived from a force plate, can determine torque about a joint. Applying gait analysis equipment to ascertain torque for the ankle joint can assess the capability of novel transtibial level prostheses . Since movement disorders can impact gait, traditional clinical gait analysis systems are applicable to the evaluation of therapy and severity of Parkinson’s disease [40, 41].
However, in general there are issues regarding the prevalent application of these clinical gait analysis systems. For example the majority of these traditional clinical gait analysis systems require a gait laboratory, which may involve a predetermined walkway, rather than an autonomous environment. Furthermore, a level of specialization and teamwork is implied for the operation of some of these gait analysis devices [1, 2, 5, 16, 17, 32].
There are implications of widespread application of gait analysis systems, therefore they may be inherently constrained based on the inferred resource limitations. Furthermore, logistic considerations are apparent, such as the distance for a patient to be transported to clinical gait laboratory. The capacity to objectively evaluate and quantify a subject’s gait in an autonomous environment, such as through wearable and wireless systems would be highly desirable.
Instrumented mechanisms have been also applied to the domain of reflex quantification, such as for the patellar tendon. The patellar tendon reflex serves a contributing role to the basis for gait. In general, an instrumented reflex quantification device applies a means for quantifying the elicitation of the tendon reflex and a means to quantify the reflex response [6, 34].
Recently developed instrumented reflex quantification systems incorporate either an instrumented reflex hammer or a motorized actuator device for evoking the reflex. The response is generally quantified through EMG sensors or instrumentation for measuring the force and torque properties of the reflex response [42, 43, 44, 45, 46, 47, 48, 49]. Other more sophisticated techniques quantify the reflex response through the application of optical motion capture systems and triaxial accelerometers with tethering [50, 51, 52, 53]. However, wireless inertial sensor have rendered tethered devices obsolete .
These applications are notably transcended in terms of utility regarding the introduction of locally wireless accelerometer systems for quantifying reflex response. The wireless system alleviates encumbrances of a tethered system. With a potential energy impact pendulum for evoking the patellar tendon reflex the reflex response can be elicited in a highly controlled manner. Furthermore, with a tandem wireless accelerometers mounted proximal to the lateral malleolus and potential energy impact pendulum the latency of the patellar tendon reflex can be derived [6, 34].
Locally wireless systems for quantification of gait, reflex response, and movement disorder, a transitional step to Network Centric Therapy
Gait and reflex response are notably associated, for which likewise the application of locally wireless systems, such as accelerometers, has been successfully applied to quantify gait and reflex response [5, 6, 17, 34]. The implementation of locally wireless systems for the quantification of reflex response progressively evolved over the course of four evolutionary phases [6, 34, 55, 56, 57, 58, 59]. Central to the development of the wireless quantified reflex device is the application of a potential energy impact pendulum constructed of aluminum with a protractor to measure the level of potential energy levels imparted to evoke the patellar tendon reflex. This quantified input system also enabled the predetermined targeting of the patellar tendon [6, 34, 56, 57, 58, 59]. Further testing and evaluation applied the use of an artificial reflex system to successfully establish the efficacy of the wireless quantified reflex device without the inherent variability of neurology foundational to the reflex [60, 61].
The progressive evolution of the wireless reflex quantification device converged upon a locally wireless accelerometer that was essentially wearable about the lateral malleolus near the ankle joint through an elastic band, such as a sock. Trial data was wirelessly conveyed to a locally situated computer for subsequent post-processing. Depending on the experimental objectives the sampling rate of the wireless accelerometers could be adjusted. With tandem wireless accelerometers mounted to the subject’s lateral malleolus and potential energy impact pendulum the reflex response and latency were quantified in a consistent, reliable, and reproducible manner [6, 34, 57, 58, 59].
Since gait is associated with reflex, the extrapolation to applying locally wireless and essentially wearable accelerometer systems to quantify gait is a logical extension of the technology’s capability [5, 6, 34]. Wearable accelerometer systems have been successfully contrasted to conventional gait analysis equipment . Preliminary wearable and wireless accelerometer systems for measuring gait consisted of integrated systems with local wireless connectivity [63, 64]. Some of these applications required highly specialized mounting techniques to achieve their desired effect [62, 63]. However, other wearable and wireless apply simplified mounting techniques, such as proximal to the ankle or the actual shoe worn during gait [64, 65].
The successful research, development, testing, and evaluation of the wireless quantified reflex device was eventually extrapolated to gait by LeMoyne et al. These research experiments applied the same wearable and locally wireless accelerometer system to quantify gait in a relatively autonomous setting [66, 67, 68]. The primary emphasis was oriented toward the evaluation of hemiplegic gait and the capacity to develop a real-time feedback rehabilitation strategy for hemiplegic gait [66, 67].
The application of accelerometer systems that are effectively wearable has also been demonstrated for the quantified evaluation of movement disorders. Preliminary research has been demonstrated for the identification of movement disorder status and response to therapy invention [69, 70, 71, 72, 73, 74]. Additionally, other techniques for quantifying movement disorder have utilized locally wired inertial sensors with connectivity to a module for local wireless transmission of the trial data .
The implementation of a wearable and wireless system has been successfully demonstrated through the same wireless accelerometer system utilized by the wireless quantified reflex device. By mounting the wireless accelerometer about the dorsum of the hand, simulated Parkinson’s disease tremor has been successfully quantified in an considerably objective manner [76, 77]. The wireless accelerometer system can be mounted to the dorsum of the hand as a wearable device through a simple glove .
The preliminary success of wearable and locally wireless systems for the quantification of gait and associated reflex quantification, and movement disorder establish the foundation for the third generational phase of Network Centric Therapy. Network Centric Therapy is prevalently demonstrated through the application of the smartphone as a wearable and wireless inertial sensor system [1, 2, 3]. Regarding the domains of gait and reflex quantification the smartphone and associated portable media device have been successfully applied for an assortment of scenarios [1, 18, 23, 24, 25, 26]. Furthermore, the smartphone has been extended to the evaluation of movement disorder status, with successfully determination of deep brain stimulation efficacy [27, 28, 29, 30].
Preliminary extension of wearable and wireless systems for the quantification of gait and movement disorder, the origins of Network Centric Therapy
During 2010 LeMoyne and Mastroianni sought to evolve beyond the successful application of wearable and locally wireless systems for the quantification of human movement. A ubiquitous observation of the smartphone is the change in screen orientation occurs upon modifying the position of the smartphone itself. The smartphone is equipped with an inertial sensor package that measures spatial orientation. With a software application the smartphone can convey an inertial sensor signal recording as an email attachment through wireless connectivity to the Internet for pending post-processing anywhere in the world [1, 2, 3, 18, 23, 24, 25, 26].
Using the smartphone as a wearable and wireless accelerometer system based on a software application LeMoyne and Mastroianni conducted a gait quantification experiment supervised by Mastroianni in the general area of Pittsburgh, Pennsylvania. The trial data for the gait quantification experiment was conveyed as an email attachment through wireless connectivity to the Internet for post-processing by LeMoyne in the metro-Los Angeles, California region. A representative illustration of the mounting of the smartphone proximal to the lateral malleolus is provided in figure 1. Figure 2 demonstrates the acquired acceleration waveform during gait .
Figure 1. Mounting of a smartphone about the lateral malleolus for the quantification of gait .
Figure 2. Acceleration waveform for gait acquired by a smartphone functioning as a wearable and wireless accelerometer system .
LeMoyne and Mastroianni determined that the smartphone with software application to function as a wearable and wireless accelerometer system could be readily applied to quantify hand tremor for Parkinson’s disease. The experiment conducted in the general area of Pittsburgh, Pennsylvania applied a smartphone mounted about the dorsum of the hand as illustrated in figure 3. The experimental contrasted a subject with and without Parkinson’s disease. The experimental data was transmitted by wireless connectivity to the Internet as an email attachment for subsequent post-processing in the metro-Los Angeles, California region. The acceleration waveform for the subject with Parkinson’s disease hand tremor is presented in figure 4. The acceleration waveform for the subject without Parkinson’s disease with a steady hand is displayed in figure 5. Statistical significance was achieved between the subject with Parkinson’s disease and without Parkinson’s disease .
Figure 3. The smartphone mounted by glove to the dorsum of the hand for quantifying Parkinson’s disease hand tremor .
Figure 4. The acceleration waveform quantifying hand tremor for the subject with Parkinson’s disease .
Figure 5. The acceleration waveform quantifying the steady hand for the subject without Parkinson’s disease .
The findings of the 2010 endeavor of LeMoyne and Mastroianni for the quantification of movement disorders establish the foundation for the concept of Network Centric Therapy. An immediate observation was that the experimental and post-processing resources could be situated anywhere in the world. In essence the most remotely situated patient could potentially have access to the best clinical resources in the world, and the response to a therapy intervention strategy could be visualized the quantified feedback of the wearable and wireless inertial sensor system. The Internet connectivity to an email resource essentially represents a functional Cloud computing resource [1, 2, 3, 19, 23, 24, 25, 26]. Subsequent research further substantiated the relevance of wearable and wireless systems for the quantification of gait and movement disorders through the use of a smartphone [78, 79, 80, 81].
During 2011 another observation was that the portable media device retains analogous functional properties as a wearable and wireless system to the smartphone with the same software application. The portable media device requires a local wireless zone to establish connectivity to the Internet. Using a similar mounting procedure about the lateral malleolus of the ankle the portable media device was successfully demonstrated to quantify gait . Another benefit of applying is that tandem activated portable media devices can quantify disparity of hemiplegic gait in a statistically significant context, which would be potentially cost prohibitive regarding the application of two smartphones . Later during 2018 LeMoyne et al. demonstrated that a single smartphone could be applied to determine statistical significance between an affected and unaffected hemiplegic leg pair during gait with statistical significance while maintaining gait velocity consistency through a treadmill .
Wearable and wireless systems for quantifying reflex response with eventual machine learning classification, another Network Centric Therapy application
Since the quantification of reflex response, such as the patellar tendon, is generally reserved to a clinical setting, the portable media device requiring a local wireless Internet zone is highly relevant for application. During 2012 the wireless quantified reflex device was modified with a portable media device for the successful acquisition of quantified patellar tendon reflex response . A further evolution of the quantification of reflex response would be the ability apply machine learning as an augmented diagnostic methodology.
A notable aspect of the quantification of reflex response is the apparent ability to distinguish the accelerometer signals regarding a hemiplegic reflex pair, such as the patellar tendon. The affected leg displays a perceptibly notable amplified reflex response. Machine learning provides a suitable means to distinguish between a hemiplegic patellar tendon reflex pair with a classification accuracy . Machine learning enables a unique diagnostic capability, especially in light of the observation that even skilled clinician have disputed the presence of an asymmetric hemiplegic reflex pair [6, 10, 34, 86].
The process for applying machine learning through, such as the Waikato Environment for Knowledge Analysis (WEKA) involves consolidating trial data provided by the inertial signal into a feature set. The process for reducing the inertial sensor signal trial data to a feature set generally applies software automation, for which an Attribute-Relation File Format (ARFF) is developed for machine learning classification [31, 32, 86, 87, 88, 89]. WEKA provides an assortment of machine learning algorithms, which is extremely valuable, since the suitability of a particular machine learning algorithm is contextual to the feature set under consideration [31, 32, 87, 88, 89, 90].
Therefore, LeMoyne et al. sought to distinguish between an affected leg and an unaffected leg for a hemiplegic patellar tendon reflex pair using a wearable and wireless inertial sensor system. The portable media device was selected instead of the smartphone, because of the localized nature of the experimental setting. An appropriate software application enabled the portable media device to function as a wearable and wireless accelerometer platform, which was amalgamated to the wireless quantified reflex device. Given the inherent nature of the feature set, the support vector machine was selected as the machine learning algorithm. Perfect classification accuracy was attained for distinguishing between the affected leg and the unaffected leg of the hemiplegic patellar tendon reflex pair .
The inertial sensor system inertial to a wearable and wireless system, such as a smartphone and portable media device, is equipped with both an accelerometer and gyroscope [1, 18, 23, 24, 25, 26]. This assortment of inertial sensors can be applied at the discretion of the research team. Especially with regards to the characteristics of the patellar tendon reflex response, the gyroscope provides a clinically intuitive signal. Using the smartphone as a smartphone as a wearable and wireless gyroscope platform integral to the wireless quantified reflex device a consistently reliable set of trial data was obtained for the quantification of the patellar tendon reflex .
Later using the same conceptual theme of integrating the wearable and wireless system the quantification of the patellar tendon reflex, the essence of Network Centric Therapy was demonstrated. A patellar tendon reflex quantification experiment was conducted in Lhasa, Tibet, and the trial data was post-processed in Flagstaff, Arizona of the United States of American effectively on the opposite side of the world. Given the nature of the accessibility to the Internet, the portable media device was selected as the wearable and wireless system. A software application enabled the gyroscope signal to be recorded and conveyed as an email attachment through connectivity to the Internet for subsequent post-processing. The findings also revealed consistently reliable set of trial data was obtained for the quantification of the patellar tendon reflex .
Further endeavors using the wireless quantified reflex device with the smartphone as a wearable and wireless gyroscope system integrated the efficacy of machine learning. Using software automation to consolidate the gyroscope signal data to a feature set, considerable classification accuracy was attained for differentiating between the patellar tendon reflex response of the affected leg and unaffected leg for a hemiplegic reflex pair . Also using supra-maximal manual stimulation of the patellar tendon reflex through a reflex hammer, considerable machine learning classification accuracy was achieved using the portable media device as a wearable and wireless gyroscope system [94, 95]. Multiple machine learning algorithms provided by WEKA have demonstrated the capacity to attain considerable classification accuracy for distinguishing between the patellar tendon reflex response for an affected leg and unaffected leg hemiplegic pair using a smartphone as wireless gyroscope platform for a wearable and wireless system .
Wearable and wireless systems for quantifying gait, with relevance to Network Centric Therapy application
The capacity to quantify gait through wearable and wireless systems, such as the portable media device and smartphone, in the context of preliminary Network Centric Therapy has been successfully demonstrated. These endeavors have progressed from evaluating health gait to identifying disparity of hemiplegic gait. Descriptive and inferential statistics have been applied during the post-processing phase as deemed appropriate for the context of the experiment [1, 18, 20, 21, 22, 23, 24, 25, 26, 82, 83, 84]. The extension to the domain of machine learning classification would further establish the diagnostic efficacy inherent to Network Centric Therapy [1, 3, 26, 31].
The gyroscope sensor facilitates gait analysis through its clinically recognizable signal. LeMoyne and Mastroianni during 2018 conducted research, development, testing, and evaluation of hemiplegic gait using a smartphone as a wearable and wireless system. A software application enabled the smartphone to function as a wireless gyroscope platform with experimental trial data wirelessly conveyed as an email attachment through connectivity to the Internet. The integration of a treadmill into the experiment ensured relatively constant velocity throughout the gait experiment, as a single smartphone functioning as a wearable and wireless system acquired the gyroscope signal of the hemiplegic affected leg and unaffected leg .
Figures 6 and 7 present the perceptibly disparate gyroscope signals of the hemiplegic affected leg and unaffected leg, respectively. Given the nature of the gyroscope signals, descriptive statistics, such as maximum, minimum, mean, standard of deviation, and coefficient of variation, compose a relevant feature set. Through the application of WEKA, a multilayer perceptron neural network achieves considerable classification accuracy. Figure 8 presents the graphic user interface of WEKA for attaining machine learning classification accuracy, and figure 9 presents the pertinent multilayer perceptron neural network .
Figure 6. Gyroscope signal of hemiplegic affected leg during gait .
Figure 7. Gyroscope signal of unaffected leg during gait .
Figure 8. The Waikato Environment for Knowledge Analysis (WEKA) graphic user interface for attaining machine learning classification accuracy .
Figure 9. Multilayer perceptron neural network for achieving considerable classification accuracy for distinguishing between a hemiplegic affected leg and unaffected leg through a smartphone functioning as a wearable and wireless gyroscope sensor system .
This demonstration of preliminary Network Centric Therapy through the application of a smartphone as a wearable and wireless system underscores the utility of overall concept [1, 97]. Multiple Network Centric Therapy strategies, such as ankle rehabilitation, Virtual Proprioception using eccentric strengthening, wobble board therapy, and analysis of reduced arm swing status can be prescribed in a rehabilitation context using Network Centric Therapy [1, 98, 99, 100, 101, 102]. The rehabilitation response to a prescribed therapy can be then evaluated using machine learning amalgamated with a wearable and wireless inertial sensor system, such as a smartphone [1, 97].
Efficacy of deep brain stimulation for treatment intervention of movement disorder using wearable and wireless systems, such as a smartphone, a pathway to Network Centric Therapy
As of 2010 LeMoyne et al. demonstrated the capacity to quantify movement disorder hand tremor, such as Parkinson’s disease . The inertial sensor signal can provide an objective quantification of the status of movement disorder tremor. With this signal data the efficacy of a therapy intervention, such as deep brain stimulation, can be established through augmented diagnostic techniques, such as machine learning [23, 24, 25, 26].
Deep brain stimulation enables a transformative therapy intervention for the treatment of movement disorders, such as Parkinson’s disease and Essential tremor. However, attaining the optimal deep brain stimulation parameter configuration can represent a daunting task for a highly skilled clinician, since there are a considerable quantity of permutations for parameters, such as amplitude, pulse width, frequency, and electrode polarity [28, 29, 30, 31, 103, 104, 105]. From the perspective of engineering proof of concept deep brain stimulation ‘On’ and ‘Off’ status have been successfully demonstrated using the smartphone as a wearable and wireless inertial sensor system, which establish a pathway for automated deep brain stimulation tuning [23, 24, 25, 26, 27, 28, 29, 30].
During 2015 LeMoyne et al. applied a smartphone as a wearable and wireless accelerometer system for the quantification of deep brain stimulation using ‘On’ and ‘Off’ status with machine learning classification. The experiment involved mounting the smartphone to the dorsum of the hand with a latex glove. In order to manifest the symptoms of Essential tremor, the subject was presented with a reach and grasp task. The experimental trial data of the accelerometer signal was conveyed as an email attachment through wireless connectivity to the Internet. The accelerometer signal was consolidated into a feature set suitable for machine learning classification using WEKA. Considerable machine learning classification through a support vector machine attained for distinguishing between deep brain stimulation ‘On’ and ‘Off’ status based on quantified feedback from a smartphone functioning as a wireless accelerometer platform . Other similar experiments using the multilayer perceptron neural network as a machine learning algorithm also achieved considerable classification accuracy .
During 2018 LeMoyne et al, applied a similar research perspective for the successful quantification and machine learning classification of deep brain stimulation for Parkinson’s disease hand tremor with the application of the smartphone as a wearable and wireless inertial sensor system. For this experimental research the smartphone was equipped with an application that could simultaneously record the accelerometer and gyroscope signal for subsequent wireless transmission to the Internet as an email attachment. The two signal increase the number of attributes for the feature set. Using a multilayer perceptron neural network considerable classification accuracy was achieved for distinguishing between deep brain stimulation ‘On’ and ‘Off’ status using a smartphone as a wearable and wireless inertial sensor system .
Another consideration for machine learning classification in the context of distinguishing appreciable classification accuracy is the time to determine the classification accuracy. Therefore, evaluating multiple machine learning algorithms in terms of both classification accuracy and time to achieve classification accuracy warrants consideration. Six machine learning algorithms through WEKA were evaluated: multilayer perceptron neural network, support vector machine, K-nearest neighbors, logistic regression, J-48 decision tree, and random forest .
This experiment also the smartphone was mounted to the dorsum of the hand as a wearable and wireless inertial sensor system to quantify Parkinson’s disease hand tremor and secured by a latex glove as illustrated in figure 10. The smartphone recorded and conveyed both the accelerometer and gyroscope signal as email attachments through wireless Internet connectivity. Figures 11 and 12 represent the accelerometer magnitude signal of Parkinson’s disease hand tremor with deep brain stimulation set to ‘On’ and ‘Off’ status, respectively .
The three machine learning algorithms achieved the highest classification accuracy for distinguishing between deep brain stimulation ‘On’ and ‘Off’ status: multilayer perceptron neural network, support vector machine, logistic regression. The respective multilayer perceptron neural network is represented in figure 13. However, support vector machine and logistic regression attained their classification accuracies in time spans that were roughly an order of magnitude less than the multilayer perceptron neural network. Therefore the support vector machine and logistic regression machine learning algorithms may be more suitable for real-time processing scenarios, especially if the processing resources are constrained .
Figure 10. Representative illustration of a wearable and wireless inertial sensor system (smartphone) mounted to the dorsum of the hand by a latex glove to quantify Parkinson’s disease hand tremor .
Figure 11. Smartphone derived accelerometer magnitude signal of Parkinson’s disease hand tremor with deep brain stimulation set to ‘On’ status .
Figure 12. Smartphone derived accelerometer magnitude signal of Parkinson’s disease hand tremor with deep brain stimulation set to ‘Off’ status .
Figure 13. Multilayer perceptron neural network for distinguishing between deep brain stimulation ‘On’ and ‘Off’ status for Parkinson’s disease .
In summary Network Centric Therapy for the treatment of movement disorders envisions the application of a wearable and wireless systems with an inertial sensor to record and wirelessly convey trial data pertaining to symptoms, such as hand tremor, to a Cloud computing environment. Expert clinical resource teams can post-process the data anywhere in the world with augmented diagnostic and prognostic strategies, such as machine learning. Even through the patient and clinical resources are remotely situated, the clinician could provide optimized medical intervention strategies and effectively a real-time optimized deep brain stimulation parameter configuration.
Future perspectives on Network Centric Therapy
Network Centric Therapy encompasses the application of wearable and wireless systems for the quantification of an assortment of human movement characteristics, such as gait and associated reflex response and movement disorders, through devices equipped with inertial sensors, such as the smartphone and portable media device. The current state of technology for Network Centric Therapy, such as enabled by the smartphone and portable media device, demonstrates the considerable potential of the Internet of Things for the domain of biomedical engineering and healthcare. Future extrapolation project the prevalence of wearable and locally wireless sensor level nodes capable of conveying their inertial sensor data to more powerful wireless systems, such as the smartphone, portable media device, and tablets [1, 2, 3, 106, 107].
Network Centric Therapy represents the transformative application of the Internet of Things to the domain of biomedical engineering and healthcare. Preliminary demonstration of the benefits of Network Centric Therapy have been provided through the use of wearable and wireless systems, such as the smartphone and portable media device, which are equiped with inertial sensors, such as the accelerometer and gyroscope. These devices have been successfully demonstrated for the quantification of gait and associated reflex, with augmented diagnostic capability through machine learning classification. Furthemore, wearable and wireless systems, such as the smartphone, have been applied to quantify movement disorders involving hand tremor. This achievement has be extrapolated to machine learning classification to distinguish between deep brain stimulation regarding ‘On’ and ‘Off’ status for movement disorders, such as Parkinson’s disease and Essential tremor, with the future objective of developing real-time parameter configuration optimization.
With wireless connectivity to the Internet the quantified experimental data can be stored in an effective Cloud computing environment for post-processing. The implication is that the patient, clinicial team, and post-processing resources can be situated anywhere in the world. Future evolutions of Network Centric Therapy envision the development of locally wireless inertial sensors with connectivity to Cloud computing resources through intermediary connectivity to smartphones, portable media devices, and tablets. This capabilty will further enhance patient autonomy. In summary, Network Centric Therapy will radically transform rehabilitation and therapy intervention relative to previous subjectively interpreted ordinal scale techniques and highly resource intensive instrumented methodolgies generally confined to clinical laboratories.