Detection and assessment of the severity of Levodopa-induced dyskinesia in patients with Parkinson's disease by neural networks

Levodopa-induced dyskinesias detection based on the complexity of involuntary movements

“…Generally, the methods of ambulatory motor assessment in PD that distinguish between ON, OFF and LID state use numerous features computed from signals in a frequency range less than 8 Hz, recorded from multiple body segments with accelerometers while the subjects perform typical daily tasks (Hoff et al, 2001;Keijsers et al, 2000Keijsers et al, , 2003Keijsers et al, , 2006. In an experiment with PD patients having mild to severe LID it was shown that the mean amplitude of the dominant frequency in two frequency bands (1-4 Hz and 4-8 Hz) correlates well with the m-AIMS scores only in the tasks in which the voluntary movements were suppressed (like sitting in a chair and abstaining from any voluntary movement) (Hoff et al, 2001).…”

“…like drinking from a cup, putting on a coat, buttoning or walking). In order to differentiate LID from the voluntary movement, the accelerometer features must be used together with trainable pattern recognition classifiers like neural networks (Keijsers et al, 2000;Keijsers et al, 2003). In a scheme based on a two-layer MLP a number of 38 parameters, computed from the accelerometer signals, were used in order to detect LIDs and quantify their severity in 16 PD patients at rest and during various activities of daily living (Keijsers et al, 2000).…”

“…In order to differentiate LID from the voluntary movement, the accelerometer features must be used together with trainable pattern recognition classifiers like neural networks (Keijsers et al, 2000;Keijsers et al, 2003). In a scheme based on a two-layer MLP a number of 38 parameters, computed from the accelerometer signals, were used in order to detect LIDs and quantify their severity in 16 PD patients at rest and during various activities of daily living (Keijsers et al, 2000). Posture and movements were measured by using eight uniaxial accelerometers attached in orthogonal pairs (tangentially and radially) to the wrist, the upper arm (just above the elbow joint), the trunk (sternum) and the leg (just above the knee joint) of the most affected side.…”

“…Posture and movements were measured by using eight uniaxial accelerometers attached in orthogonal pairs (tangentially and radially) to the wrist, the upper arm (just above the elbow joint), the trunk (sternum) and the leg (just above the knee joint) of the most affected side. There is an important difference between our detection method and the method from Keijsers et al, 2000 (other than using different kind of input features): we had a classification problem while in that approach the goal was to optimize the automatic assessment of LID. More exactly, while we trained our MLPs to recognize from input features one of the two classes (NonDPD/DPD) the MLPs of Keijsers et al were trained to predict the m-AIMS score.…”

A holistic approach to assess the feasability of telehealth for Parkinson's disease management

“…Parkinson's disease patients showed overall less adaptations in relative phase than the control group, especially those involving the upper extremities [336]. In The Netherlands, Keijsers et al carried out several research works using four pairs of triaxial accelerometers mounted on the wrist, upper arm, trunk, and leg on the most affected side in combination with neural networks to detect and quantify the dyskinesia and LID as well as to study the motor fluctuations in PD patients [337]- [340].…”