Martyna Michałowska

The Recognition of Human by the Dynamic Determinants of the Gait with Use of ANN

In this paper human recognition method is based on dynamic parameters of the human gait. In the method the artificial neural network algorithm is employed. Some parameters of the gait are defined in order to describe and recognize characteristics of gait for each considered individual. In classical approach to human recognition, description of the gait pattern is based on sequence of images analysis and biometric parameters. In this paper we present method based on determining some dynamic characteristics of the gait, which together with other kinematic determinants should allow to describe the unique gait pattern for each individual. Necessary data were obtained from system of motion analysis BTS and force plates, commonly used in biomechanics. All considered gait parameters were calculated from data which were obtained for 15 persons with different characteristic of the gait. To implement the recognition process the back-propagation neural network algorithm was used. In the paper three configurations of the input data (only kinematic parameters from the BTS system, only dynamic parameters from the force plates and both types of parameters together) are investigated and compared.

Discussion about different cut-off values of conventional hamstring-to-quadriceps ratio used in hamstring injury prediction among professional male football players

Objective To measure the sensitivity and specificity of differences cut-off values for isokinetic Hcon/Qcon ratio in order to improve the capacity to evaluate (retrospectively) the injury of hamstring muscles in professional soccer screened with knee isokinetic tests. Design Retrospective study. Methods Medical and biomechanical data of professional football players playing for the same team for at least one season between 2010 and 2016 were analysed. Hamstring strain injury cases and the reports generated via isokinetic testing were investigated. Isokinetic concentric(con) hamstring(H) and quadriceps(Q) absolute strength in addition with Hcon/Qcon ratio were examined for the injured versus uninjured limbs among injured players, and for the injured and non-injured players. 2 x 2 contingency table was used for comparing variables: predicted injured or predicted uninjured with actual injured or actual uninjured. Sensitivity, specificity, accuracy, positive and negative predictive values, and positive and negative likelihood ratio were calculated for three different cut-off values (0.47 vs. 0.6 vs. 0.658) to compare the discriminative power of an isokinetic test, whilst examining the key value of Hcon/Qcon ratio which may indicate the highest level of ability to predispose a player to injury. McNemar’s chi2 test with Yates’s correction was used to determine agreement between the tests. PQStat software was used for all statistical analysis, and an alpha level of p <0.05 was used for all statistical comparisons. Results 340 isokinetic test reports on both limbs of 66 professional soccer players were analysed. Eleven players suffered hamstring injuries during the analysed period. None of these players sustained recurrence of hamstring injury. One player sustained hamstring strain injury on both legs, thus the total number of injuries was 12. Application of different cut-off values for Hcon/Qcon significantly affected the sensitivity and specificity of isokinetic test used as a tool for muscle injury detection. The use of 0.47 of Hcon/Qcon as a discriminate value resulted in significantly lower sensitivity when compared to 0.658 threshold (sensitivity of 16.7% vs. 91.7%, respectively; t = 6.125,p = 0.0133). Calculated values of specificity (when three different cut-off were applied) were also significantly different. Threshold of 0.6 of Hcon/Qcon resulted with significantly lower specificity compared to 0.47 value (specificity of 46.9% vs. 94.5%, respectively; t = 153.0,p<0.0001), and significantly higher specificity when compared to 0.658 (specificity of 46.9% vs. 24.1%, respectively; t = 229.0, p<0.0001). Conclusion The use of different cut-off values for Hcon/Qcon significantly affected the sensitivity and specificity of isokinetic testing. The interpretation of usefulness of isokinetic test as a screening tool in a group of male professional football players to predict hamstring injury occurrence within the next 12 months might be therefore significantly biased due to the different threshold values of Hcon/Qcon. Using one “normative” value as a cut-off (e.g. 0.47 or 0.60, or 0.658) to quantify soccer players (or not) to the group with a higher risk of knee injury might result in biased outcomes due to the natural strength asymmetry that is observed within the group of soccer players.

Gender recognition using artificial neural networks and data coming from force plates

The paper deals with a problem of automatic gender recognition based on parameters obtained from the force plates. The ground reaction force is recorded and some selected parameters of the curve are calculated. These parameters are used in this study as inputs to artificial neural network which should recognize if the individual is male or famale. The results of recognition are satisfactory and presented in the paper.

On Different Methods for Calculating the Flight Height in the Vertical Countermovement Jump Analysis

Vertical countermovement jump is a very simple and common method for assessing the jumping ability of athletes. There are different techniques for measurements of the flight height, e.g. using motion capture systems or accelerometers. In this paper for estimating the flight height the measurements coming from the force plates are used. Furthermore different methods can be applied for calculating the flight height based on these measurements. In these paper four methods of calculating the flight height during the vertical countermovement jump based on the measurements from the force plates are compared (the flight time method, the take-off velocity method, the work-energy method and the center of jumper’s body vertical position method). In addition for two of these methods (the take-off velocity method and the center of jumper’s body vertical position method) the authors applied two different methods of numerical integration (the trapezoidal rule and based on the cubic spline interpolation).

Application of Artificial Neural Networks in the Human Identification Based on Thermal Image of Hands

The aim of this study was to check the possibility of identifying the persons based on the properties of thermal maps and a temperature distribution of a hand, obtained from a thermal image, with use of artificial neural networks. For this purpose, a series of thermographs of the right hand of eight people was taken, with a thermal imaging camera. The photos were taken under the same thermal conditions, but with different state of warming of hands. After processing the photos (determining the edges, characteristic hand points and areas of interest), the parameters characterizing the metacarpal temperature distribution were determined. Eight parameters were chosen, which were average temperatures of the areas of interest. These parameters were input data of neural networks in the learning and identification process. As it was shown in this study, these parameters were sufficient to clearly identify the persons. Neural networks, designed as multi-layered perceptron, after proper learning showed very high values of identification parameters, including high values of sensitivity and specificity, what proves the high quality of classification. Such identification is possible with the natural thermal state of the hand and if thermal images are not strongly disturbed, the artificial neural networks are very good tool to implement in persons identification process.

Comparison of Some Evolutionary Algorithms For Optimization of the Path Synthesis Problem

The paper presents comparison of the results obtained in a mechanism synthesis by means of some selected evolutionary algorithms. The optimization problem considered in the paper as an example is the dimensional synthesis of the path generating four-bar mechanism. In order to solve this problem, three different artificial intelligence algorithms are employed in this study.The paper presents comparison of the results obtained in a mechanism synthesis by means of some selected evolutionary algorithms. The optimization problem considered in the paper as an example is the dimensional synthesis of the path generating four-bar mechanism. In order to solve this problem, three different artificial intelligence algorithms are employed in this study.

Artificial neural networks in knee injury risk evaluation among professional football players

Lower limb injury risk assessment was proposed, based on isokinetic examination that is a part of standard athlete’s biomechanical evaluation performed mainly twice a year. Information about non-contact knee injury (or lack of the injury) sustained within twelve months after isokinetic test, confirmed in USG were verified. Three the most common types of football injuries were taken into consideration: anterior cruciate ligament (ACL) rupture, hamstring and quadriceps muscles injuries. 22 parameters, obtained from isokinetic tests were divided into 4 groups and used as input parameters of five feedforward artificial neural networks (ANNs). The 5th group consisted of all considered parameters. The networks were trained with the use of Levenberg-Marquardt backpropagation algorithm to return value close to 1 for the sets of parameters corresponding injury event and close to 0 for parameters with no injury recorded within 6 - 12 months after isokinetic test. Results of this study shows that ANN might be useful tools, which simplify process of simultaneous interpretation of many numerical parameters, but the most important factor that significantly influence the results is database used for ANN training.Lower limb injury risk assessment was proposed, based on isokinetic examination that is a part of standard athlete’s biomechanical evaluation performed mainly twice a year. Information about non-contact knee injury (or lack of the injury) sustained within twelve months after isokinetic test, confirmed in USG were verified. Three the most common types of football injuries were taken into consideration: anterior cruciate ligament (ACL) rupture, hamstring and quadriceps muscles injuries. 22 parameters, obtained from isokinetic tests were divided into 4 groups and used as input parameters of five feedforward artificial neural networks (ANNs). The 5th group consisted of all considered parameters. The networks were trained with the use of Levenberg-Marquardt backpropagation algorithm to return value close to 1 for the sets of parameters corresponding injury event and close to 0 for parameters with no injury recorded within 6 - 12 months after isokinetic test. Results of this study shows that ANN might be useful...

Flow of subretinal liquid through the retinal hole after surgery – mechanical model and FEM simulations

The objective of this study is to analyze subretinal liquid dynamics by means of numerical simulations in patients with retinal detachment after insertion of intussusception. The main parameter being analyzed is the outflow velocity of subretinal fluid through the hole in the retina. All calculations were performed using FEM and the results were evaluated in relation to the velocity of outflow when considering parameters such as; i) speed of eyeball movement, ii) shift of intussusception in relation to the center of the hole in the retina, iii) viscosity of the subretinal fluid. Results obtained suggest an optimal speed of eyeball movement for patients to achieve maximal speed of outflow of subretinal fluid. This is a significant factor in achieving improved clinical outcomes for patients.

HAMSTRING/QUADRICEPS RATIO IN ACL INJURY PREDICTION IN ELITE FOOTBALL PLAYERS

Background Hamstring/quadriceps conventional ratio (H/Q) has been widely used in the prediction of anterior cruciate ligament (ACL) injury; however there is still no consensus on the values that are really predictive for ACL injury in elite male football players. Objective To verify the value of H/Q predictive for ACL injury in elite male footballers. Design Retrospective study, analysing ACL injury occurrence in footballers and H/Q value from isokinetic tests. Setting Male football elite division. Patients (or Participants) Records of 74 professional male footballers who played at least one season between 2010–2016 have been analysed. Interventions (or Assessment of Risk Factors) All players were monitored and treated by one clinic, thus medical history of each player was documented. ACL injury was confirmed in USG examination. Players were dichotomously categorized as injured or uninjured for data analysis. The same clinic was responsible for conducting isokinetic assessment using Biodex System 3 Pro. Retrospective footballers records were reviewed between May 2006 and April 2016. Receiver operating characteristic (ROC) curves were used to evaluate the possibility of using the H/Q value to predict ACL injury. Level of significance was set at p<0.05 and Statistica v.12.5 software was used. Main Outcome Measurements Area under the curve (AUC); number of ACL injuries and values of H/Q were also analysed. Results Total 331 knee isokinetic tests of 74 elite male football players were reviewed, and data of 8 ACL injuries were analysed. Cut-off value of H/Q ratio was calculated at 0.63±0.11. The ROC analyses showed an area under the curve of 0.67 (95% CI 0.49–0.84; p=0.0598). Conclusions Although it occurred to be insignificant we observed a tendency towards H/Q values at 0.63±0.11, that may be useful in future studies in multifactorial analysis of ACL injury prediction. However, studies on bigger sample should be performed to verify these results.

APPRAISAL OF THE FUNCTIONAL MOVEMENT SCREEN IN FOOTBALL INJURY PREDICTION

Background Functional Movement Screen (FMS) has been used in the prediction of football injuries. However, the discussion continues on how sensitive and specific the test is. Lately, systematic review [McCall et al. 2014] enrolling studies published up to September 2014 and clinical commentary [Wright et al. 2016] were published underlying caution prior to implementation of the FMS into clinical practice as an injury prediction tool. Thus, it seems necessary to up-date the debate. Objective To provide up-to-date evidence-based summary of available evidence regarding the predictive value of FMS in football players. Design Critically Appraised Topic. Setting Amateur, collegiate Division I and II. Patients (or Participants) Male and female footballers. Interventions (or Assessment of Risk Factors) Medline, SPORTDiscuss/with/Full Text, Academic Search Complete, Nursing/Academic Edition, MasterFILE Premier databases were searched (on August 31st 2016) using “football”, “soccer”, “functional movement screen” keywords. Only prospective cohort studies published as full article in reviewed journals were included. Main Outcome Measurements Between samples – injured vs. uninjured analysis, odd ratio (OR), likelihood ratio (LR), receiver operating curve (ROC). Results Our search yielded 18 studies, after abstract revision 3 articles met the criteria. The first [Warren et al. 2015] did not confirm the validity of FMS in injury prediction (OR=1.01, 95% CI 0.53–1.91). The second [Schroeder et al. 2016] showed no difference between injured/uninjured players in FMS score (Z=−0.890, p=0.373), concluding FMS to be a poor predictor of injury in football players. The third [Mokha et al. 2016] found that asymmetry was a better predictor of injury than the composite FMS score. None of the studies used LR ratio or ROC curve analyses in statistical analysis. Conclusions Caution is required when applying FMS in football injury prediction. It is necessary to conduct a well-designed prospective study with sufficient statistical analysis used for predictive parameters to justify the use of FMS test in injury prediction in football players.