PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 88 NR 12b/2012 103 Adam ŚWITOŃSKI 1,2 , Magdalena STAWARZ 1 , Magdalena BOCZARSKA-JEDYNAK 4 , Aleksander SIEROŃ 3 , Andrzej POLAŃSKI 1,2 , Konrad WOJCIECHOWSKI 1,2 Polish-Japanese Institute of Information Technology (1), Silesian University of Technology, Institute of Computer Science (2) Medical University of Silesia, Department of Internal Diseases, Angiology and Physical Medicine (3), Department of Neurorehabilitation and Department of Neurology, Medical University of Silesia, Central University Hospital (4) The effectiveness of applied treatment in Parkinson disease based on feature selection of motion activities Streszczenie. W pracy zaprezentowano analizę skuteczności stymulacji prądowej jądra niskowzgórzowego oraz leczenia farmakologicznego w chorobie Parkinsona. W tym celu badano właściwości dyskryminacyjne, charakterystycznych cech ruchu. Dla danych kinematycznych przeprowadzono ekstrakcję, a następnie selekcję cech. Zastosowano ranking atrybutów bazujący na entropii i zachłanne przeszukiwanie wspinaczkowe z oceną średniej odległości wewnątrzgrupowej. Uzyskane wyniki wykazują większy wpływ stymulacji prądowej na badane czynności ruchowe. (Skuteczność leczenia w chorobie Parkinsona na podstawie selekcji charakterystycznych cech ruchu). Abstract. The analysis of effectiveness of deep brain stimulation and pharmacological treatment in Parkinson disease is presented. It is based on an examination of discriminative properties of distinctive motion features. The feature extraction and selection of kinematical motion data is carried out. The attribute ranking with entropy based attribute evaluation and greedy hill climbing search with assessment of an average inner class dissimilarity are applied. The obtained results show that deep brain stimulation has greater impact on investigated motion activities. Słowa kluczowe: choroba Parkinsona, pomiar ruchu, selekcja cech, ekstrakcja cech, nadzorowane uczenie maszynowe Keywords: Parkinson diseases, motion capture, feature selection, feature extraction, supervised machine learning Introduction Parkinson’s disease (PD) is a chronic progressive disease, which belongs to the group of motor system disorders. The dysfunction of movement are caused by the loss of dopamine-producing brain cells in the substantia nigra, a region located in the midbrain. Nowadays only symptomatic methods of treatment have been applied, because the reason of cells destruction in substantia nigra is not known. The main motor features of PD are: tremor in hands, arms, legs, jaw and face, rigidity of the limbs and trunk, bradykinesia, impaired balance and coordination. The primary pharmacological drug is L-dopa, a specific amino acid, which after reaching the brain, is converted into dopamine. Fundamental pharmacological medication with L-dopa is successful for first years, but later the progression of neurodegeneration as well as dopaminergic treatment itself results in motor complications. There is an alternative symptomatic treatment for advanced PD patients using Deep Brain Stimulation (DBS). DBS of the subthalamic nucleus (STN) has become an established therapy for patient with PD, it is an effective and safe method of symptomatic treatment of PD patients, who are medically resistant to pharmacotherapy [1][2]. The objective assessment of the treatment carried out by clinicians is based mainly on the Unified Parkinson’s Disease Rating Scale (UPDRS). The motor part of UPDRS consist of 14 points, which evaluate different motor skills based on discrete scale in range of 0-4, where 0 means normal ability to move. In this study we compare and demonstrate possibilities of the developed multi-featured MOCAP measurement system on medical examination data of the Parkinson Disease patient who has undergone the surgery based on implanting Deep Brain Stimulator for improving his motoric skills. The patients taking part in this research were operated on in the Department of Neurosurgery Medical University of Silesia in Katowice[8][9][10]. All measurements were done in multimodal Human Motion Laboratory of Polish – Japanese Institute of Information Technology (PJWSTK) in Bytom, Poland. The laboratory allows to acquire motion data through simultaneous and synchronous measurement and recording of motion kinematics, muscle potentials by electromyography, ground reaction forces and video streams in high definition format Data were collected from PD patients during four experimental conditions called sessions, defined by pharmacological medication and subthalamic nucleus electrical stimulation: Session1: StimOFF/MedOFF, Session2: StimON/MedOFF, Session3: StimOFF/MedON and Session4: StimON/MedON. Experimental scenario includes seven tasks and has been planned based on criteria taking into consideration in motor examination part of UPDRS scale. In the described work data from Task 2 and Task 4 are used. Task 2 contains gait measurements, performed across a straight line with different speed. In Task 4, a pull test is carried out. A participant is standing erect on the platform with feet no more than shoulder width apart and is pulled back. It allows to asses the ability to recovery on his own and evaluate the postural stability.. Fig. 1 Markers tracked by 10-camera, 3D motion capture system (Vicon). In Fig. 1 locations of 39 attached markers on a human body, tracked by the motion capture cameras are presented. The assumed skeleton model, reconstructed on the basis of markers positions contains 24 segments with rotations coded by Euler angles. There are also global rotation and translation data included in the kinematical frame description, which results in 78 dimensional pose space. What is more, muscle potentials of the lower body parts are captured by 16 electrodes of EMG subsystem and ground reaction forces are measured by two plates as shown in Fig. 1.