- |
-
Application of Radial Basis Function Artificial Neural Network in Diagnosing Athletes Anterior Cruciate Ligament Rupture
Oral Presentation
Paper ID : 1432-11THCONF
Authors
1master student of mechanical engineering University of Tabriz ,
2Assistant Professor of sport biomechanics engineering , Sport Sciences Research Institute Of Iran
3Associate Professor of mechanical engineering , University of Tabriz
4PhD student, Faculty of Sport Science, Tehran University, Tehran, Iran
Abstract
Introduction: The cruciate ligament rupture is one of the most common threatening injuries for athletes. If the recovery process and exercises to strengthen, not done properly after surgery, the risk of injury After returning to sport will be high. So, feature extracting and identifying the features of emg signals for athletics who have acl rupture and healthy athletics can be used to diagnose anterior cruciate ligament status of who have no information about a history of surgery or damage in his cruciate ligament area. The purpose of this research is to extract features and classification between healthy and surgryed acl by generalized regression neural network, which is a kind of radial basis function artificial neural networks.
Methodology: In this research, surface emg signals has been recorded from eight muscles that are effective in knee area during the landing test from platform on target foot. (Gluteus Medius ،Vastus Medialis ،Vastus Lateralis، Adductor Magnus ،Lateral Hamestering ،Medial Hamestering ،Lateral Gastrocnemius، Medial Gastrocnemius). The athletes under test included two groups of patients with anterior cruciate ligament rupture that lasted at least 6 months after surgery and healthy athletes. The landing test carried out 3 times for each person and the energy of emg signals for each person, is calculated by Pseudo Wigner Ville distribution. then، the main features of each signal is extracted by using marginal properties of pseudo wigner-ville distribution after normalizing. The marginal properties of landing tests 1 and 2 have been used to train/create a two-layer generalized regression neural network. marginal features of third landing test is used to test and obtaining the ability of network in this issue.
Results: By using the designed neural network and testing non trained data's, this network is able to classify 67% of the unknown signals correctly.
Discussion: In the other words ، by creating/training the designed neural network, the values of weights and biases of network have changed in such a way that the network can detect the health status of ACL by comparing the pseudo wigner-ville distribution features of lower limb muscles emg signals with what the network learned in training process for a person whose ACL status is unknown.
Methodology: In this research, surface emg signals has been recorded from eight muscles that are effective in knee area during the landing test from platform on target foot. (Gluteus Medius ،Vastus Medialis ،Vastus Lateralis، Adductor Magnus ،Lateral Hamestering ،Medial Hamestering ،Lateral Gastrocnemius، Medial Gastrocnemius). The athletes under test included two groups of patients with anterior cruciate ligament rupture that lasted at least 6 months after surgery and healthy athletes. The landing test carried out 3 times for each person and the energy of emg signals for each person, is calculated by Pseudo Wigner Ville distribution. then، the main features of each signal is extracted by using marginal properties of pseudo wigner-ville distribution after normalizing. The marginal properties of landing tests 1 and 2 have been used to train/create a two-layer generalized regression neural network. marginal features of third landing test is used to test and obtaining the ability of network in this issue.
Results: By using the designed neural network and testing non trained data's, this network is able to classify 67% of the unknown signals correctly.
Discussion: In the other words ، by creating/training the designed neural network, the values of weights and biases of network have changed in such a way that the network can detect the health status of ACL by comparing the pseudo wigner-ville distribution features of lower limb muscles emg signals with what the network learned in training process for a person whose ACL status is unknown.
Keywords
Subjects