Page 74 - ASME DSCC 2015 Program

Technical Program

Combining Genetic Algorithms and Extended Kalman filter to Estimate Performance Evaluation of a Sensorized Arthroscopic Grasper
Ankle’s Muscle-tendon Parameters Contributed short paper. DSCC2015-9946
Contributed regular paper. DSCC2015-9781 Behnaz Poursartip, Daniel Yurkewich, Marie-Eve leBel, Rajni V. Patel,
Luis enrique coronado, raúl chávez romero, Mauro e. Maya, antonio Ana luisa Trejos, Michael naish, Western University, London, ON, Canada
Cardenas, Universidad Autonoma de San Luis Potosí, San Luis Potosí, SLP, Force sensing minimally invasive instruments have gained increasing atten-
Mexico, Davide Piovesan, Gannon University, Erie, PA, United States
tion in recent years. Integrating these instruments within currently available
This work proposes a set of simulation and experimental measurements to surgical simulators can enhance the learning experience by measuring
estimate muscle biomechanical parameter during human quiet standing. the forces applied by trainees and supplementing objective performance as-
Understanding the mechanisms involved in postural stability is sessment. Recently, an arthroscopic grasper was designed and sensorized
indispensable to improve the knowledge of how humans can regain with Fiber Bragg Grating Sensors at Canadian Surgical Technologies and
balance against possible disturbances. Postural stability requires the Advanced Robotics (CSTAR). Moreover, a custom low-cost (LC) interrogation
ability to compensate the movement of the body’s center of gravity system was developed to accompany the proposed sensorized tool. In this
caused by external or internal perturbations. This paper describes the study, the custom LC interrogator was used and compared to the commer-
implementation of a hybrid parameter-estimation approach to infer the cially available Micron Optics sm130 (MO) interrogator. The hypothesis is
features of the human neuro-mechanical system during quiet standing and that both of these systems can be used to measure forces within ±0:5 N as
the recovery from a fall. The estimation techniques combines a genetic the acceptable margin for accuracy. Experimental results showed that the
algorithm with the State-Augmented Extended Kalman Filter. These two MO system meets the required accuracy for certain force directions. The LC
algorithms running sequentially are utilized to estimate the musculo- system demonstrated 49% of the accuracy of the MO interrogator. The main
skeletal parameters. advantage of the LC interrogator is its cost, which is 18% of the commercial
This paper shows results of the approach when representing human stand- interrogation system. For certain force directions, the performance was
ing as either a second-order or third order mechanical model. Experimental comparable to the defined criteria.
validation on a human subject is also presented.
Prediction of Periventricular leukomalacia occurrence in neonates
Meal Detection and Meal Size Estimation for Type 1 Diabetes using a Novel Support Vector Machine classifier optimization Method
Treatment: A Variable State Dimension Approach Invited session paper. DSCC2015-9984
Contributed regular paper. DSCC2015-9905
Dieter Bender, Ali Jalali, C. nataraj, Villanova University, Villanova,
jinyu xie, Pennsylvania State University, State College, PA, United States, PA, United States, Daniel licht, Children’s Hospital of Philadelphia,
Qian Wang, Pennsylvania State University, University Park, PA, United Philadelphia, PA, United States
States
Prior work has documented that Support Vector Machine (SVM) classifiers
To compensate the glucose variability caused by meals is essential in can be powerful tools in predicting clinical outcomes of complex diseases
developing Artificial Pancreas for type 1 diabetes. Most existing algorithms such as Periventricular Leukomalacia (PVL). Our previous study showed
rely on meal announcements and determine the insulin doses based on that SVM performance can be improved significantly by optimizing the
an Insulin-to-Carbohydrate ratio (I:C ratio). However, patients, especially supervised training set used during the learning stage of the overall SVM
young patients, often forget to provide meal information under natural living algorithm. This study fully develops the initial idea using the reliable
conditions. A Variable State Dimension (VSD) based algorithm is developed Leave-One-Out Cross-validation (LOOCV) technique. The work presented
to detect meals which are not announced by patients. The algorithm is eval- in this paper confirms previous results and improves the performance of the
uated using an FDA-approved UVa/Padova simulator and has demonstrated SVM even further. In addition, using the LOOCV technique, the computation-
to achieve 95% success rate in meal detection with less than 17% false alarm al time is decreased and the structure of the algorithm simplified, making
rate. In addition, the average meal size estimation error is no more than 13%. this framework more feasible. Furthermore, we evaluate the performance of
We then integrate the VSD-based meal detection and estimation algorithm the resulting optimized SVM classifier on an unseen set of data. This demon-
with our previous published glucose dynamics model consisting of both in- strates that the developed SVM algorithm outperforms normal SVM type
sulin and carbohydrate inputs. The goodness of fit for 30min-ahead glucose classifiers without any loss of generalization.
predictions using meal information provided by the VSD-based algorithm
has increased by 86% in average compared to the prediction using a model
without meal input based on plasma blood glucose (BG) data. Simulation
results also show that compared to several meal detection/estimation algo-
rithms in the literature, the VSD-based algorithm has comparable or shorter
detection time.

74