Automatic Classification of Gait Patterns in Cerebral Palsy Patients

The application of wearable sensors coupled with diagnostic models presents one of the most recent advancements in automation applied to the medical field, allowing for faster and more reliable diagnosis of patients. Nonetheless, such applications pose a complex challenge for traditional intelligent automation (combining automation and artificial intelligence) methods due to high class imbalances, the small number of subjects, and the high dimensionality of the measured data streams. Furthermore, automatic diagnostic models must also be explainable, meaning that medical professionals can understand the reasoning behind a predicted diagnosis. This paper proposes an intelligent automation approach to the diagnosis of cerebral palsy patients using multiple kinetic and kinematic sensors that record gait pattern characteristics. The proposed artificial intelligence framework is a multi-view fuzzy rule-based ensemble architecture, in which the high dimensionality of the sensor data streams is handled by multiple fuzzy classifiers and the high class imbalance is handled by a cost-sensitive training algorithm for estimating a fuzzy rule-based stack model. The proposed methodology is first tested on benchmark datasets, where it is shown to outperform comparable benchmark methods. The ensemble architecture is then tested on the cerebral palsy dataset and shown to outperform comparable ensemble architectures, particularly on minority class predictive performance.