Neural network models for customized alignment of endoskeleton BK prosthesis

Resulting from various medical conditions, natural disasters and human conflicts numerous lower limb amputations are performed annually. A large number of these patients can stand, walk, run and climb with the aid of a prosthesis which consists of a foot, a shank/pylon, a custom designed socket, and a pair of alignment devices at either end of the shank. The alignment devices allow for optimal positioning of the artificial foot relative to the socket and the residual limb. Optimal alignment ultimately determines the comfort, stability, suspension, energy conservation of the prosthesis. Hence, sub-optimal alignment, even with a perfect fitting socket, may lead to instability and excessive energy consumption, resulting in fatigue and skin breakdown. In this research, we exploit the underlying relationship between the demographics of the patients and their gait patterns. A neural network with a back propogation architecture is trained by a random optimization algorithm. The alignments suggested by the trained neural network are validated against the final dynamic alignment done by a pool of expert prosthetists. In an effort to maintain integrity and reliability of the neural network model, the validation data is independent and separate from the training data sets.