The Neurorehabilitation Workstation is described. The need to maintain a clinical perspective motivates the comprehensive nature of the system, which integrates multiple data acquisition devices, interface technologies, advanced analytical techniques, and multi-sensory rendering capabilities. Emphasis is placed on machine-resident intelligence embedded at several levels.

Introduction

The field of Rehabilitation applies techniques and resources from many disciplines and is constantly seeking to improve the measurement of human performance and the assessment of therapeutic efficacy. We have had considerable success recently in our attempts to transfer new technologies into the clinical setting for such purposes. Devices such as gloves to measure hand motion dynamics, surface EOG and EMG sensors for eye movement and muscle contraction, and lightweight pressure sensor arrays for gait analysis show great promise in therapy. At the same time, our efforts to make these transfers permanent have been impeded by the lack of standard platforms, interfaces, inaccessible file formats, as well as the medical community's lack of time, technical expertise, and adequate budgets. Until now no cost-effective solution appeared possible. Recent developments in human-computer interface hardware and software, data analysis, and expert systems suggest this is no longer the case. We are currently exploring a solution, the Neurorehabilitation Workstation (NRW), which integrates these technologies and methods into a comprehensive system designed specifically for the clinic. In addition, we hope it may be generic enough to act as a standard for other similar applications. The success of the NRW depends on four things; modular design (for distributed processing and adaptability), integration of several data input devices into a single platform within a common interface protocol, implementation of machine-resident intelligence (neural nets, fuzzy logic) on several levels, and creation of a development environment driven by clinical needs. We detail aspects of these features below.

Data Input

A necessary feature of the NRW is the integration of a variety of data input devices into a single system to include EEG, EMG, EOG, ECG, dynamic bend sensors, pressure sensors, audio and video digitizers, etc. The resulting capacity for data fusion allows for meaningful correlations to be made across various performance modalities. The devices and their hardware boards connect to an external module, and a high speed bus will route the data both to a central multi-tasking server and to the rendering subsystem for immediate feedback. The server should be intelligent enough to automatically implement a custom configuration of input device parameters, interface functionality, and relevant records based on the device(s) connected and the identity of the operator(s) and patient(s) currently at the system.

Data Management

The maintenance of medical record integrity is a significant issue. Such integrity is achieved through security protocols, standardized data formats, error handling, and semi-automated database archiving. The data management subsystem tasks also include linking the device data with the patient record and specifying sensor-specific data formats and structures.

Interactive Modalities/Methodologies

The user interface will be based on new theories of human-computer interaction methodologies , computer-supported cooperative work, knowledge engineering, expert systems, and adaptive task analysis The system will monitor a user's actions, learn from them, and adapt by varying aspects of the system's configuration to optimize performance. Adaptable on-line knowledge-based help using text, graphics, and animated tutorials provide interactive learning and navigation.

Data Analysis

Effective therapeutic intervention relies on a comparative evaluation of a patient's progressing or digressing state. The nature of the change in this state may often be quite subtle, even imperceptible using traditional techniques. Given that the data acquisition subsystem can detect these changes, the data analysis subsystem is designed to enhance them in ways that may then be rendered to optimize the operator's sensory modalities. Linear and nonlinear multivariate analysis tools will be capable of processing multiple data sets in a variety of ways, including graphical analysis (phase portraits, compressed arrays, recurrence maps, etc.) and sound editing (mixing, filtering). Automated detection of trends and correlations using fuzzy logic may be performed in the background or in a post-processing mode. The user may then be alerted by the system if it detects areas worthy of further investigation. Such a feature should expedite the creation of a taxonomy of lesion-specific impairments.

User Classification

We have defined five types of users. These types help define discrete levels of user functionality. Therapists, the primary users of the system, are responsible for data acquisition, data management, basic analysis, and patient-oriented interactive biofeedback modes. Technicians are responsible for simple data acquisition. Physicians will use the more comprehensive data analysis tools. Researchers will focus on the data analysis but their use of the system will be unconstrained. They will explore and develop custom analytical techniques. Patients will primarily use the therapeutic biofeedback features of the system, usually in a supervised setting.

Data Rendering Modalities

With multi-sensor data acquisition and advanced analytical characterization, the rendering capacity of the system becomes extremely vital. The NRW will implement multi-sensory rendering by combining recently developed 3D sound and tactile feedback systems with advanced visualization technologies. Research in human sensory physiology has shown the eye to be optimized for feature extraction of spatially-rendered data, the ear for temporally-rendered data, and the tactile sense for textures [9]. Thus the NRW will enhance perception of complex relationships by integrating visual, binaural, and tactile modalities. The rendering subsystem has a near real-time biofeedback mode for use in a therapeutic paradigm and a data perceptualization mode for use in an analytical paradigm. Outputs from sensing devices and analytical operations are parsed and routed to the combination of rendering modalities best suited to render that information.

Conclusion

The goals of the NRW are twofold; 1) to provide an open hardware platform and modular infrastructure which will expedite the implementation of new technologies into the clinic, 2) to augment clinical therapy with new methods of interaction and analysis. Success should result in providing neurorehabilitation, and the medical community in general, with a powerful tool for characterizing the complex nature of normal and impaired human performance.