TITLE: An Integrated Approach to the Design of Goal-Oriented Dynamic Networks.

ABSTRACT: The presentation shall provide an introduction to the concepts and principles that are being used to design dynamic networks capable of controlling inertial objects.  These networks consist of a collection of interconnected nodes or units that exhibit time-varying behavior.  The individual units possess "intrinsic" behavioral goals that may be pursued regardless of desired "global" network goals specified by a supervisory agent.  Additionally, network units may or may not be willing to consider the well-being of fellow network units as they make decisions regarding current and future actions.  The following questions are being addressed:

1. Can we develop a general "template" for the goal-oriented dynamic networks under consideration?
2. What are the fundamental concepts that should be considered when deciding how the network units are connected, or coupled, together?
3. Are there patterns of network connectivity that allow the system to achieve both unit-level and network-level goals?
4. How do we "tune" these networks?
5. What is the structure of the "tuned" dynamic networks that are capable of achieving desired measures of performance? Can we quantify this structure?
6. Do these results provide any insight into the known structure and performance of biological nervous systems?

The present work draws upon concepts from many fields of study including, but not limited to, organization theory, control theory, dynamic optimization, neuroscience, artificial neural networks,
statistical learning, and graph theory.

Abstract: We proposed a multi-step signal processing method to extract  the possible fetal EEG component from transabdominal recordings. Our study shows that the method yields good quantitative and subjective results of fetal EEG extraction.

Introduction 
Cerebral palsy patients exhibit a wide range of functional responses to specific surgical interventions.  The variability associated with surgical outcomes within this patient population precipitates the need to determine the functional effects in advance of the surgery.  This study seeks to determine the accuracy and stability of neural network models for the pre diction of  post-surgical kinematics. Pilot studies have indicated that the application of neural networks toward predicting functional outcomes of surgery produces clinically meaningful results, and that creation and implementation of neural networks provides the surgeon with important information
during the pre-surgical decision-making process. 

Methods 
Post-surgical kinematic outcomes were predicted using a variety of neural network and statistical techniques.  These methods included the Ward Backprop neural network, shape analysis with locally weighted regression, and the Cascade Correlation neural network.  Preliminary studies were completed using the Ward Backprop networks with twenty three legs.  However, this network did not prove to be as accurate as the Cascade Correlation network, and so it was not used in the final analysis.  Another study used 100 legs to compare Cascade Correlation networks with locally weighted regression, a statistical matching and prediction technique.  Finally a new method that used distance metrics to pre-sort data for inclusion in the neural network training set was evaluated.  The similarity of the predictions to the post-surgical data was evaluated using statistical similarity measures including a distance metric, and a clinical similarity statistic.  This latter statistic was derived from a web based survey of gait clinicians. 

Results and Discussion 
 The neural network model was found to extrapolate predictions more accurately than shape analysis with locally weighted regression.  Results indicated that it was possible to predict 60% of the post-surgical waveforms for the hip and 75% of post-surgical waveforms for the knee using the neural network compared to 35% and 60% respectively using locally weighted regression.   Thirty percent of neural network predictions were within the average daily variation of knee flexion found by Miller et al. (1996).  Where prediction was not possible, pre-surgical characteristics that may confound the predictive models were identified.  These included subjects who did not exhibit impact absorption during support phase, and subjects with crouched gait.  In addition, hamstring lengthening was determined to be predictive of rectus transfer outcomes.  Finally, the new pre-sorting technique appeared to be on a par or better than earlier techniques in predicting post-surgical outcomes. 

     Arthroplasty (TKA) reliably reduces pain and improves function in patients with knee osteoarthritis, but quadriceps strength and function lag behind healthy, age-matched adults years after surgery.  The cause of persistent quadriceps weakness after TKA is unclear, but it may be related to deficits in quadriceps muscle activation. Current rehabilitation does not typically address muscle activation deficits. The purpose of this study was to quantify the loss of quadriceps strength and activation 3-4 weeks after TKA and determine the effectiveness of adding neuromuscular electrical stimulation (NMES) to rehabilitation for patients with unilateral and bilateral TKAs. 
     Quadriceps strength and activation were tested in twenty patients before and 3-4 weeks after unilateral TKA. Thirty-four patients with unilateral TKA participated in a randomized clinical trial examining the effectiveness of adding NMES to a 6 week rehabilitation program beginning 3 weeks after TKA. Quadriceps strength, muscle activation, functional performance, perceived health statuses were tested 3, 6, 9, and 12 weeks after TKA. Eight patients with bilateral TKA participated in a similar study with NMES.
     Patients 3-4 weeks after TKA had a profound loss of quadriceps strength (62%) and activation (16%).  The decrease in activation explained 65% of the strength loss. Pain during a maximal voluntary quadriceps contraction explained 39% of the muscle activation deficit. 
In the randomized clinical trial for patients with unilateral TKA, patients who had NMES included in rehabilitation had greater improvements in muscle activation within the first 3 weeks of treatment, with a similar trend for quadriceps strength.  Patients who received NMES also had significantly greater improvements in functional performance during Timed-Get-Up-and-Go and Functional Stair Tests at various test intervals.  There were no meaningful differences between treatment groups for functional performance (Timed-Get-Up-and-Go and Functional Stair Tests) or health status measures (SF-36, KOS, and GRS).  Patients after bilateral TKA had the greatest increase in quadriceps strength when NMES was added to rehabilitation (431%) compared to patients who received only voluntary exercise (180%). 
     These results suggest that NMES has considerable potential for improving quadriceps strength and countering muscle activation deficits, as well as for decreasing the length of rehabilitation for patients after TKA.