Multimodal Behavioural Assessment in Rehabilitation

Abstract

After suffering from a serious injury, illness or surgery, the patient usually needs to follow a long and critical physical rehabilitation program to recover the former strength, mobility and fitness. Procedures for monitoring patients' movements are widely used in this context and are mainly aimed at identifying and maximizing life quality and movement potential. Many rehabilitation programs rely on classical treatments based on physiotherapy, which requires trained specialists and their precious experience. However, sometimes, these treatments lack standardized and objective information for properly evaluating patients' performance. Creating non-invasive systems, using a low-power scheme at low cost, for monitoring the patients' movements and interpreting properly the data acquired in order to provide them with a useful feedback, is one of the current challenges and the main aim of this PhD dissertation. As a contribution to the assessment of rehabilitation exercises, a hardware and software suite is proposed, both for human motion acquisition and tracking and data analysis.

Firstly, to capture human motion, a prototypical wearable sensing system, based on a robust communication scheme and a data alignment algorithm, is developed and tested. It is composed of a number of small modules that embed high-precision accelerometers and wireless communications to transmit the information related to the body motion to a workstation. The system includes a device for video acquisition during the training sessions in order to provide, additionally, visual feedback to the patient.

Afterwards, three different methods are investigated for the analysis of motion data. The first one is based on neural gas networks, which are considered in a implementation of the Memory Prediction Framework theory. The Granular Linguistic Model of a Phenomenon, which is based on the Computational Theory of Perceptions, is used in combination with a Fuzzy Finite State Machine in the second approach to develop a tool that analyses the exercise and provides a linguistic description of it. The last method proposes a hybrid neuro-fuzzy system which merges the two previous schemes in order to join their advantages while compensating their drawbacks.

From the experimental results obtained, it can be concluded that, both from a practical and a theoretical point of view, the proposed suite can introduce improvements in the current state of the art in the field of human motion monitoring for rehabilitation.