[7A5] Kohonen neural network analysis of postural sway in stand-still position in healthy adult subjects
O D Ojie and R Saatchi
Sheffield Hallam University, UK
A prototype accelerometry device that recorded and analysed balance information to assist clinicians in diagnosing vestibular dysfunction was developed. It was evaluated on 20 healthy adult volunteers and the results demonstrated that it can provide effective measurement of the body's sway path. Sway path balance data were recorded from the subjects using the accelerometry device in accordance with the modified Clinical Test for Sensory Interaction of Balance (mCTSIB). This test examines human balance in four scenarios: (i) eyes open standing on a firm surface; (ii) eyes closed standing on a firm surface; (iii) eyes open standing on a flexible surface (such as a foam); and (iv) eyes closed standing on a flexible surface. The accelerometry data were converted to balance parameters in the medio-lateral (ML) and anterior-posterior (AP) directions, namely: rms distance-ML, rms distance-AP, rms velocity-ML, rms velocity-AP, rms acceleration-ML, rms acceleration-AP, range of distance-ML, range of distance-AP, range of velocity-ML, range of velocity-AP, range of acceleration-ML, range of acceleration-AP, average distance-ML, average distance-AP, average velocity-ML, average velocity-AP, average acceleration-ML, average acceleration-AP. Using paired sample t-test, the parameters that were able to differentiate between at least one of the possible six combinations of the tests (ie condition 1 against 2, 1 against 3, 1 against 4, 2 against 3 , 2 against 4 and 3 against 4) were chosen. These were root mean square velocity and acceleration, medio-lateral (ML) and anterior and posterior directions, range of velocity and acceleration, anterior-posterior (AP), average velocity and average acceleration, range of distance AP and ML and rms velocity-ML. The Kohonen network was used to explore the effectiveness of these parameters in clustering sensory information based on mCTSIB. The Kohonen neural network analysis was repeated when its inputs were transformed by principal component analysis (PCA) and results were compared with those obtained without the application of PCA to determine whether PCA improved the analysis. The tests indicated the suitability of the chosen variables for balance analysis and the manner the four tests associated with mCTSIB related in healthy adults.
