Objective: Application of a computerised algorithm to kinematic motor recordings to distinguish parkinsonian bradykinesia from normal movement.

Background: Bradykinesia, the only obligatory physical sign for a diagnosis of Parkinson’s disease, is a shorthand for complex disturbances of initiation and execution of actions and the ability to sustain them. This makes it difficult to identify reliably, particularly in the early stages of the disease.

Methods: Cartesian Genetic Programming (CGP) is an evolutionary algorithm used to train and validate classifiers of a data stream. We applied this to movement recordings of the three upper limb bradykinesia tasks from the MDS-UPDRS motor subscale: finger tapping, hand pronation-supination and hand opening-closing. Twenty-two Parkinson’s disease patients and 20 controls were studied with an electromagnetic motion tracking system. The inputs to the computation were based on the characteristics of bradykinesia according to the MDS-UPDRS-III—speed and size of movement, rhythmicity, decremental tendency. These motor features were adapted to the specific requirements of each task. A classifier was defined as the evolved algorithm, after 10,000 ‘generations’ at a ‘mutation’ rate of 0.05. Three sets of classifiers were evolved, one for each movement task. The fitness assigned to each classifier was the proportion of samples correctly classified.

Results: For the finger tapping task, averaged accuracy for the best classifier was 82.66%; for the pronation-supination task, 80.54%; and for the hand opening-closing, 75.32%. Using the model’s ability to recognise which inputs were used to evolve the strongest classifier, the most discriminating motor features for each task were derived.

Conclusions: CGP can be applied to all three of the bradykinesia tasks of the MDS-UPDRS-III scale to develop effective classifiers. By applying the algorithm to a larger patient sample, the overall accuracy could be increased and there would be a more robust description of the most discriminative movement features. CGP has the capacity to accept raw positional or speed data points and to perform unbiased searching, which is not constrained by pre-defined characteristics. Such use of raw data to induce classifiers may better recognise the fundamental defects in the execution of motor plans that are contained in the term parkinsonian bradykinesia. Data presented at BIOSIGNALS conference 19-21 January 2018.

« Back to 2018 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/evaluation-of-bradykinesia-in-parkinsons-disease-using-a-computerised-evolutionary-algorithm/