Computer aided diagnosis of low back disorders using the motion profile

Abstract

Feature extraction and stepwise discrimination techniques were applied to motion profiles (MPs) obtained from subjects during a repetitive trunk flexion and extension task, in order to develop a computer aided procedure for the diagnosis of low back disorder. 524 subjects, belonging to normal, low back pain patient, and pre-employment categories were tested using the B-200 Isostation dynamometers, at the Mayo Clinic and The Ohio State University. Principal components analysis (PCA) and Fourier descriptor (FD) methods were used to efficiently represent the continuous MPs and phase portraits, respectively, by reducing the dimensionality of the data. In addition, the statistical parameters from the continuous MPs were used to represent the dynamic trunk performance. The results of discriminant analysis indicated similar error rates ranging from 19% to 24%, using the three methods of data representation: MPs, PCA and FD.