Artificial Neural Networks (ANN) Applied for Gait Classification and Physiotherapy Monitoring in Post Stroke Patients

1.1 Overview of the problem Humans have an innate predisposition for ambulation (walking). The motor neuron stimulation involved in ambulation is generated by a natural neural network located in the spinal cord, known as the central pattern generator for locomotion. This network is strongly influenced both by super-spinal structures situated mainly in the hypothalamus and brainstem, and by signals coming from various types of peripheral receptors (Carter & Page 2009). To facilitate research and analysis, free gait in humans is traditionally divided into phases and cycles. Each full gait cycle comprises two individual steps; a single step consists of a stance phase and a swing phase. The gait cycle includes a stage of single limb stance (when the body rests on a single lower extremity) and a double limb stance (on both lower extremities). Kinematic gait analysis assumes a simplified, 15-segment model of the human body (feet, shins, thighs, forearms, upper arms, hands, head, torso, and pelvis) (Blaszczyk 2004). There are two kinds of basic parameters adopted for gait modelling and routine testing of ambulation in healthy and disabled individuals: spatial values of motion (including step length, velocity of the body mass centre, progressions of changes in joint angles, body mass oscillations) and dynamic values of gait mechanics (most often including ground reaction forces in 3 planes and the distribution of foot forces on the ground). These physical values are measured in parallel with bioelectric muscle activity (EMC), registered by surface electrodes as a subject walks (Perry & Burnfield, 2010). Correct ambulation requires the precise integration of practically all the systems of the human body. When one of the elements, especially a motor organ, is damaged as a consequence of injury, degeneration, or deformation, this immediately finds reflection in divergences of the above parameters from normative values, which is in practice described as pathological gait (Perry & Burnfield, 2010). The field of clinical biomechanics therefore