Analysing functional measurements and questionnaire results for automated back-pain disorder classification

There is a strong need to develop a system that objectively and automatically interprets examination re-sults for people with potential back-pain problems. The right suggestion for a suitable exercise or reha-bilitation program in an early phase instead of having a patient make a number of consultations and ex-aminations can save large amounts of money to the healthcare system. Much of the knowledge used to make a correct diagnosis is implicitly present in the clinician's experience. Therefore it has proven to be difficult to use straightforward rules or algorithms to automate this task. In this study we aimed to de-velop classifiers using physical-measurement oriented data sets as well data consisting of answers to questionnaires concerning lifestyle, subjective experiences and emotional states. Statistical analyses were performed to find information shared between variables and reduce the feature set. Linear dis-criminant classifiers, logistic regression classifiers and artificial neural networks were used to develop a system that is able to indicate whether or not a person would need further examination because of poten-tial back-pain problems. The problem appears to be too complex for a linear approach, non-linear ap-proaches like artificial neural networks and logistic regression classifiers prove to be suitable for this problem.