Provided for non-commercial research and educational use only. Not for reproduction, distribution or commercial use. This chapter was originally published in the book International Review of Neurobiology, Vol. 107 published by Elsevier, and the attached copy is provided by Elsevier for the author's benefit and for the benefit of the author's institution, for non- commercial research and educational use including without limitation use in instruction at your institution, sending it to specific colleagues who know you, and providing a copy to your institution’s administrator. All other uses, reproduction and distribution, including without limitation commercial reprints, selling or licensing copies or access, or posting on open internet sites, your personal or institution’s website or repository, are prohibited. For exceptions, permission may be sought for such use through Elsevier's permissions site at: http://www.elsevier.com/locate/permissionusematerial From: Charles W. Lu, Parag G. Patil and Cynthia A. Chestek, Current Challenges to the Clinical Translation of Brain Machine Interface Technology. In Clement Hamani and Elena Moro, editors: International Review of Neurobiology, Vol. 107, Burlington: Academic Press, 2012, pp. 137-160. ISBN: 978-0-12-404706-8 © Copyright 2012 Elsevier Inc. Academic Press Author's personal copy CHAPTER SEVEN Current Challenges to the Clinical Translation of Brain Machine Interface Technology Charles W. Lu*, Parag G. Patil*,†,1, Cynthia A. Chestek* *Department of Biomedical Engineering, University of Michigan, Ann Arbor, Michigan, USA †Department of Neurosurgery, University of Michigan Health System, Ann Arbor, Michigan, USA 1Corresponding author: e-mail address: [email protected] Contents 1. Introduction and Motivation 137 2. Device Hardware 141 3. Performance 147 4. Surgical Considerations 149 5. Conclusion 152 References 152 Abstract Development of neural prostheses over the past few decades has produced a number of clinically relevant brain–machine interfaces (BMIs), such as the cochlear prostheses and deep brain stimulators. Current research pursues the restoration of communication or motor function to individuals with neurological disorders. Efforts in the field, such as the BrainGate trials, have already demonstrated that such interfaces can enable humans to effectively control external devices with neural signals. However, a number of signifi- cant issues regarding BMI performance, device capabilities, and surgery must be resolved before clinical use of BMI technology can become widespread. This chapter reviews challenges to clinical translation and discusses potential solutions that have been reported in recent literature, with focuses on hardware reliability, state-of-the-art decoding algorithms, and surgical considerations during implantation. 1. INTRODUCTION AND MOTIVATION Neural prosthetic devices have the potential to provide treatment for many different neurological conditions. As this technology matures, it becomes increasingly relevant to a number of clinical applications. “Input” devices that stimulate the nervous system, in particular, have been especially International Review of Neurobiology, Volume 107 # 2012 Elsevier Inc. 137 ISSN 0074-7742 All rights reserved. http://dx.doi.org/10.1016/B978-0-12-404706-8.00008-5 Author's personal copy 138 Charles W. Lu et al. successful in achieving therapeutic effects. Cochlear prostheses, for example, have been implanted in over a hundred thousand people, enabling many profoundly deaf individuals to comprehend spoken language (Fayad & Elmiyeh, 2009). Similarly, deep brain stimulators (DBSs) are the preferred surgical treatment for late-stage Parkinson’s disease (Benabid, Chabardes, Mitrofanis, & Pollak, 2009). However, there are also many potential appli- cations for output systems, which record and decode neural signals from the brain. Output systems are commonly used to localize seizures in cases of intractable epilepsy (Kelly & Chung, 2011) and for target guidance of DBS electrodes (Chen et al., 2006; Snellings, Sagher, Anderson, & Aldridge, 2009). In recent years, research has focused on recording neural signals directly from motor cortex to predict motor movements in order to restore movement to individuals with paralysis or amputation. Paralysis is a widespread problem, common to many disorders such as spi- nal cord injury, muscular dystrophy, stroke, cerebral palsy, and amyotrophic lateral sclerosis (ALS) (Donoghue, Nurmikko, Black, & Hochberg, 2007). According to the Christopher Reeve Foundation, nearly 6 million individuals in the United States are affected by some form of paralysis, including over a million who are victims of spinal cord injury. Figure 7.1 shows a breakdown of the various causes. Older studies have estimated that there are closer to a quarter million spinal cord injury victims in the United States (Lasfargues, Custis, Morrone, Cars well, & Nguyen, 1995). In either case, however, this represents a large potential demand for implantable brain–machine interfaces (BMIs). The average annual cost of treatment per affected individual ranges from $40,000/year, for incomplete motor function, to over $170,000/year, for high tetraplegia (National Spinal Cord Injury Statistical Center, 2012). The use of neural interfaces to bypass dysfunctional neural pathways may sig- nificantly offset such costs and improve patients’ lives by restoring their ability to interact with their surroundings, via improved communication, control of a motor prosthesis, or other assistive technologies. In a 2008 survey of spinal cord injury patients (Anderson, Friden, & Lieber, 2009), 44% of respondents indicated willingness to undergo tendon transfer surgery to obtain a 30% in- crease in elbow-extension strength and 66% indicated willingness to gain a 50% improvement in pinch function. Such a procedure can be associated with a long recovery time of 6–12 weeks (Gupta, 2011; Wolford & Stevao, 2003). A BMI system may be able to provide similar or greater benefit with a less invasive surgical procedure. Anumberofneuralsignalsourceshavebeenusedtogeneratemotorcontrol signals, most notably electroencephalography (EEG), electrocorticography Author's personal copy Current Challenges to the Clinical Translation of Brain Machine Interface Technology 139 Causes of paralysis7 N = 5,596,000 Other 526,000 9% Postpolio syndrome 272,000 5% Stroke Cerebral palsy 1,608,000 29% 412,000 7% Neurofibromatosis 212,000 4% Unspecified birth defect Traumatic brain 110,000 2% injury 242,000 4% Multiple sclerosis 939,000 17% Spinal cord injury 1,275,000 23% Figure 7.1 Causes of paralysis in the United States. N 5,596,000. Reeve Foundation. ¼ Paralysis facts and figures. (ECoG), and intracortical microelectrode arrays (Schwartz, 2004; Wolpaw, Birbaumer, McFarland, Pfurtscheller, & Vaughan, 2002). All three methods typically record from the cerebral cortex, due to its accessibility and high involvement in both motor and sensory processes. EEG and ECoG detect summed field potentials via surface electrodes, placed on the scalp and cortical surface, respectively (Schwartz, Cui, Weber, & Moran, 2006). EEG systems have been able to reach a communication speed of 0.5 bps (Wolpaw et al., 2002), which may be useful to patients who are otherwise unable to communicate, except through eye movements (Krusienski et al., 2006). ECoG signals in able-bodied epilepsy patients have shown promise decoding 1 of 5 finger movements with an accuracy of 80% (Kuba´nek, Miller, Ojemann, Wolpaw, & Schalk, 2009; Wang et al., 2009). Future higher density ECoG arrays may provide even better performance (Kellis et al., 2010). However, since neural signals fall off in magnitude as the distance to the neurons increases, the highest theoretical information density can be obtained by recording from many individual neurons in the motor cortex, using penetrating high impedance electrode arrays. The exposed tips Author's personal copy 140 Charles W. Lu et al. of these wires can be positioned near individual neurons, providing high spatial and temporal resolution and allowing for single-unit recordings, as shown in Fig. 7.2.ThisreviewfocusesonBMIsutilizing penetrating microelectrode arrays, though similar device and performance challenges exist for other signal sources as well. Correlation between single-unit activity and motor control was established as early as 1966, by Evarts, who found that certain arm move- ments in monkeys could be predicted from the firing rate of individual py- ramidal tract neurons, measured using an intracortical electrode. Four years later, Humphrey, Schmidt, and Thompson (1970) showed that accurate, real-time predictions of wrist movements could be made using a small pop- ulation of recorded cells. In the following decades, development of im- proved decoding algorithms and recording techniques led to animal models capable of controlling external devices, such as a computer cursor (Carmena et al., 2003; Musallam, Corneil, Greger, Scherberger, & Andersen, 2004; Serruya, Hatsopoulos, Paninski, Fellows, & Donoghue, 2002; Taylor, Tillery, & Schwartz, 2002). The first report of a neural- controlled robotic arm was published by Chapin, Moxon, Markowitz, and Nicolelis (1999), in which rats were successfully trained to position a robotic arm in one dimension to obtain water. Similar results were produced in monkeys 1 year later, by Wessberg et al. (2000), achieving Figure 7.2 Single-unit recordings from the human temporal cortex, 5 days postimplant (Normann et al., 2009). Recordings